Hindawi

Complexity
Volume 2021, Article ID 5511866, 26 pages
https://doi.org/10.1155/2021/5511866

Review Article
Emerging Technologies of Natural Language-Enabled Chatbots: A
Review and Trend Forecast Using Intelligent Ontology
Extraction and Patent Analytics

Min-Hua Chao , Amy J. C. Trappey , and Chun-Ting Wu

Department of Industrial Engineering and Engineering Management, National Tsing Hua University, Hsinchu, Taiwan

Correspondence should be addressed to Min-Hua Chao; [email protected]

Received 19 February 2021; Revised 2 May 2021; Accepted 14 May 2021; Published 25 May 2021

Academic Editor: Abd E.I.-Baset Hassanien

Copyright © 2021 Min-Hua Chao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Natural language processing (NLP) is a critical part of the digital transformation. NLP enables user-friendly interactions between
machine and human by making computers understand human languages. Intelligent chatbot is an essential application of NLP to
allow understanding of users’ utterance and responding in understandable sentences for specific applications simulating human-
to-human conversations and interactions for problem solving or Q&As. This research studies emerging technologies for NLP-
enabled intelligent chatbot development using a systematic patent analytic approach. Some intelligent text-mining techniques are
applied, including document term frequency analysis for key terminology extractions, clustering method for identifying the
subdomains, and Latent Dirichlet Allocation for finding the key topics of patent set. This research utilizes the Derwent Innovation
database as the main source for global intelligent chatbot patent retrievals.

1. Introduction enable company organizations to answer and take care

simple questions and requested tasks from call centers, help
Despite the global impact of COVID-19, almost 80% of desks, and service agents, and at the same time pass more
global artificial intelligence (AI) projects have maintained complex issues to the real staffs and personnel, thereby
the same or even increasing the investments of R&D since controlling the human resource costs. Chatbots can save up
the beginning of the pandemic. AI-based systems nowadays to 30% of customer support costs with shortened response
are widely adopted for decision makings, which have a time and answering up to 80% of regular questions [4].
profound impact on individuals and society. The so-called The applications of intelligent chatbots have increased
intelligent systems are mostly driven by machine learning rapidly in recent years. A lot of research delves into the
(ML) or deep learning (DL) algorithms with their models details of AI and DL algorithms for chatbot solutions and
being trained and tested by big data [1]. As an important applications in pursuits of high efficiency and intelligence.
application of AI technologies, smart chatbots (or called Even though the development of chatbot seems to be
intelligent chatbots) help answer a large number of questions booming, thorough review of the life cycle of chatbot de-
related to the pandemic [2]. Statistics provide reliable in- velopments and key technologies are in great needs. Fur-
sights into trends in the intelligent chatbot development. thermore, with the popularity of the Internet and social
Reported by Business Insider, the market size of chatbots is platforms, a digitally transformed environment for the uses
expected to grow from US$2.6 billion in 2021 to US$9.4 of smart chatbots (as human machine communication in-
billion in 2024, with a compound annual growth rate terfaces) has become largely popular. More and more ap-
(CAGR) of near 30% [3]. The study shows that more than plications offer “life” services by mounting voice-interactive
50% of customers, in various business sectors, expect assistants; that is, smart chatbots, which hold regular con-
businesses to be open 24/7. Chatbots, or virtual agents, versations and provide online services interactively with
2 Complexity

users, are becoming a trend [5]. As a technology pioneer or the system, select the corresponding function, find the
market leader, based on the comprehensive review, one can corresponding file searching through complex file folders,
identify innovative technologies or applications to maintain and then finally access the needed information. With a
its lead. As a company that wants to follow the trend of chatbot, a single verbal request can complete the task.
digitization and entering into chatbot applications, betting Among enterprise-level applications, there are few voice-
resources on the most valuable development and finding the enabled chatbots, but the demand for such functions is
right breakthrough are the best strategy, knowing the state- increasing. In addition, on the premise of satisfying basic
of-the-art technologies and applications through the review service functions, soft functions are essential to the success
article. To bridge the research gap, this research aims to use of chatbots. Chatbots that incorporate features such as tone,
the intelligent ontology extraction and patent-mining emotion, and personality are desirable. Furthermore, smart
methodology to review comprehensive chatbot-related chatbots tolerate human errors or allow fuzzy requests, still
patents and their innovative technologies and applications. generate accurate answers, and are very attractive [11].
Chatbot is a computer program that allows computers to NLP technology is an important branch of AI. It studies
mimic human communications and conversations. At first, the use of computer software, such as machine learning
chatbot can only answer standard questions where questions (ML), to intelligently process natural language. The basic
and answers are known and saved in the system. With the NLP technology is mainly developed around seven levels of
technological advances, computers can gradually answer a language, including phonemes (language pronunciation
freelanced question like human by passing a Turing Test, patterns), morphology (words, how do letters form words,
which is closer to a human intelligence [6]. With the rapid the morphological changes of words), vocabulary (the re-
development of AI in recent years, intelligent chatbot has lationship between words), syntax (how words form sen-
entered a new era and has been widely applied in many tences), semantics (the corresponding meaning of language
industries. For example, the voice customer query interfaces expression), pragmatics (semantic interpretation in different
of large shopping malls, bank chatbots for monthly account contexts), and chapter (how sentences are combined into
balance queries, and even the well-known Siri reflect how paragraphs).
chatbot technology gradually enters into people’s daily lives As AI drives the transformation of the digital economy,
through intelligent interfaces. NLP is becoming the norm for companies should also pay more attention to intellectual
obtaining information, allowing companies to easily obtain property (IP) innovation and management. Therefore, it is
key information from text documents, thereby enhancing expected that the latest trend of chatbot development can be
operational efficiency or improving service levels. NLP also found from collective patent information. Through the
has many applications in other fields. Taking the medical patent layout (or landscape), important technology devel-
industry as an example, NLP technology detects signs of opment trends can be evaluated, and the development di-
cognitive impairment by analyzing the conversations be- rection of important international manufacturers can be
tween the elderly and patients with Alzheimer [7]. In the found, and international technology benchmarks can be
banking industry, optical character recognition (OCR) and used as a reference for subsequent R&D investment deci-
NLP technologies are used to automatically capture key sions [12].
document text and perform document content reviews to According to statistics from the World Intellectual
speed up the lending process [8]. For the catering services, Property Organization (WIPO), more than 80% of emerging
NLP is used to analyze customers’ comments and emotions technologies with commercial values are patented, which
for improving services or performing precision marketing shows that the patent database consists of comprehensive
[9]. domain knowledge. The purpose of the patent database is
NLP-enabled chatbot is a complex system. Starting from not only to provide a search for prior arts, but also to obtain a
the front-end user inputting utterance, the natural language wealth of information for future R&D. For example, when
understanding (NLU) module of chatbot judges the user’s key patents are found, the technology development trends
intent from the user’s natural language expression. Next, the can be extrapolated, the technical contents of domain pat-
dialogue management module finds contents that can an- ents can be analyzed, and the core countries, assignees, and
swer the user’s request. In this process, different types of inventors of the key technologies can be identified. By
databases may be accessed for finding answers. Finally, the making good use of such patent information, companies can
natural language generation (NLG) module converts the develop various business and management strategies [13].
collected contents into human-readable expression as the In order to understand the latest emerging technologies
response to the user [10]. NLP-enabled chatbot is also a of chatbots, this study takes “natural language-enabled
smart system that integrates many AI technologies. The chatbots” as the domain for relevant patent technology
chatbot technology that uses AI to imitate human conver- exploration. Thus, the overall chatbot technological devel-
sations has begun to mature and provides accurate solutions opment trends can be discovered and future research di-
or answers to complex questions. Because natural language- rections can be suggested.
enabled chatbots have the ability to map oral or written Before investigating natural language-enabled chatbots,
inputs to intent, they become popular in many applications, a well-constructed knowledge ontology is needed. After-
such as in manufacturing or service industry. Before there wards, the global patent management landscape map and
were chatbots, when employees wanted to obtain data from technology function matrix are presented. After that, a
the company’s information system, they needed to log into discussion of the analytical results will be presented to show
Complexity 3

the interesting technology trends we found and verified with review, then keyword identification and ontology gener-
the matching literature. In this study, some text-mining tools ation are carried out. The method cross-references a large
are used, such as clustering and topic modeling. Saura [14] number of technical articles and essential patents, en-
summarized the types of 11 analysis methods of data sci- suring a high coverage of technical information in specific
ences (DS) in digital marketing and provided good support fields. Finally, with text-mining technology, LDA topic
for the patent-mining analysis method used in this study. modeling method, and TFM, a complete research flow
structure is formed.
2. Literature Review
2.1. Patent Review Workflow. Past patent reviews are usually 2.2. Patent Database. In a knowledge-based economy, the
analyzed by experts. However, with the increasing number economic status of a country depends on the production,
of patents and the development of information technology distribution, and use of knowledge and information. The
[15], most patent reviews are now performed through text- latest trend of economic growth in various countries mainly
mining technology. Even with the assistance of text-mining depends on the individual’s innovative technological
technology, if there is no systematic patent review workflow, knowledge, which is an important reason why intellectual
it is likely to cause the deviations from the subject of patent property has attracted attention. Information related to
analysis. Abbas et al. [16] present an overview of the research intellectual assets, such as technical insight and legal status,
workflow and tools on patent analysis. They divide the cannot be obtained from any other literature search except
patent review workflow into three parts, including pre- for the patent database. Thus, the importance of the patent
processing, processing, and postprocessing. Retrieving database can be revealed [21]. Krejcar et al. [22] compared
patents and transformation into structured data are for several common large-scale patent databases, including
preprocessing. Extraction of structures is for processing, AcclaimIP, Symphony Innovate, Inteum, IPzen, Founda-
including key term extraction and specific statistical data. tionIP, Thomson IP Manager, and Derwent Innovation (DI),
Patent analysis approaches are for postprocessing that is and pointed out the power of DI. The DI database uses the
classified into two categories, text-mining-based approach scientific literature, global patent data, and commercial data,
and visualization approach [17]. Kim and Bae [18] present a so it can make more confident decisions in IP. Powerful
method for forecasting emerging technology of health care analysis functions and simple workflow tools make DI be the
by patent analysis. They define the patent review workflow best solution.
that can be divided into four stages, domain patent acqui- Derwent World Patents Index (DWPI) and the smart
sition, technology clustering, technology defining, and search function are two major features of DI. DWPI is a
evaluating patent clusters. They also mention that tech- process of translation, rewriting of key abstracts, content
nology clustering results may vary depending on the analyst. debugging, and normalization of patent holders after experts
In order to avoid a lack of objectivity, they focus on Co- have read the entire official patent disclosure materials,
operative Patent Classification (CPC) for forecasting which is considered to be the essence of the patent content.
emerging technology. In terms of the nonpatent literature, it The DWPI rewritten items include novelty, use, advantage,
is also an ample source of analyzing emerging technologies. technical focus, detailed description, drawing description,
Thilakaratne et al. [19] present a literature-based research activity, and mechanism. Every operation of DI simulta-
workflow. They define the article retrieval process in detail neously searches the official patent publications and DWPI
for avoiding missing any related articles, including the re- patent value-added database to obtain more complete re-
trieval rules and the standards of selection. In the article sults. This is also the unique feature of DI. Smart search will
retrieval process, defining main research purposes, key analyze the word string semantically and automatically
words, and searching strategies are three parameters for expand keywords, and then go through multiple steps of
determining the patent database. After constructing the calculation, including weighting of classification numbers
patent database, they use systematic criteria to determine and weighting of citations, to find patents related to the
the literature is relevant or not. There have three stages for input technical description. Grammar is not that important
filtering the literature: the first stage is analyzing title and here, because smart search will remove conjunctions,
abstract, the second stage is analyzing introduction and prepositions, etc. in the description and only retain the
conclusion, and the last stage is a complete reading and technical keyword description. Therefore, whether the words
using a quality checklist. After that, visualization tech- used in the technical description are accurate or whether
niques are used to present their findings. In summary, the they are mixed with too many unnecessary technical con-
entire patent review workflow can be summarized into ditions have more influence on the search results than the
three main parts, patent search for determining the da- grammar. If the keywords left by smart search after ana-
tabase, patent analysis for extracting key information, and lyzing the string are not as expected, or the results found
result display for presenting the result in an easily un- by smart search for the first time do not meet the re-
derstandable way. quirements, manually adjust, including adding new key-
Govindarajan et al. [20] proposed a systematic research words in the search pane, or removing possible noise to let
flow for industrial immersive technology. Start by iden- smart search recalculate new results. After several adjust-
tifying the domain definition and confirming the scope of ments, the result of smart search will be closer to the de-
the research, and then after the main domain technical mand. Smart search is an iterative process, the purpose is
4 Complexity

to quickly find potential targets, and if you want to search all adaptive cluster number determination method is adopted
related patents without omission, it is suitable to use general based on silhouette score. Recently, ML methods for patent
patent search technique [23]. analysis have also begun to appear. Li et al. [33] proposed a
DeepPatent that combines the convolutional neural network
(CNN) model with the word embedding model for classi-
2.3. Ontology Construction. An ontology map for a specific
fying patents. Lee and Hsiang [34] fine-tuned a bidirectional
domain connects the relevant subjects and key terms,
encoder representations from transformers (BERT) model
provides a domain knowledge-rich structure that can be as
to classify patents and compared the fine-tuned model with
the basis for analyzing technologies in depth. Weng et al.
the previously mentioned model, DeepPatent, and the result
[24] presented a lexicon-based ontology construction
shows that the precision is 9% higher. Jun [35] proposed a
method, which utilized term frequency and weighted factor
method for technical integration and analysis using boosting
to identify the relationship between key terms. If a term has
(an ML algorithm that can be used to reduce bias in su-
significant weight, then it will be imported to the lexical
pervised learning) and ensemble learning. This method uses
database. The critical words for constructing the ontology
regression trees, random forests, extreme gradient en-
are selected from the lexical database. Trappey et al. [25]
hancement, and ensemble models After analyzing the in-
proposed an information extraction approach and a
tegrated patent data, it can be extended to technology
knowledge-based ontology construction method for smart
integration and analysis in more than three technical fields.
retailing technology mining, in which unsupervised ML
methods are applied, including clustering and Latent
Dirichlet Allocation (LDA), to construct a complete on- 2.5. Technology Function Matrix (TFM). To further focus on
tology by continuously refining. Tsatsou et al. [26] proposed the patent development context of a specific technical field
an automatically constructing ontology method, which and find a technical minefield or a technical blue ocean zone,
utilized the term frequency-inverse document frequency it is necessary to analyze the technical location and function
(TF-IDF) technique to determine key terms that may be of each patent through a more detailed TFM, and further
branches or nodes of the ontology. Subhashini and Aki- explore in-depth strategies, such as technological innovation
landeswari [27] mentioned that constructing an ontology is or avoiding development conflicts [36]. In the patent
required to follow the six key steps, determining the scope of
analysis of cyber-physical systems (CPSs) and Industry 4.0,
the ontology, capturing related data, encoding those useful
Trappey et al. [37] adopted domain ontology and Interna-
data to machine-usable, integrating the results, evaluating tional Patent Classification (IPC) as the basis of TFM.
the results, and documenting the ontology. In summary, However, IPC and Cooperative Patent Classification (CPC)
constructing an ontology can mainly be divided into three are general classifications. When exploring technology in a
parts, data source, determining the relationship between specific field, a large number of patent documents may have
terms, and effectiveness evaluation. the same or similar classification codes, which makes the
identification of technical classifications insufficient, and
2.4. Patent Mining. Patent documents contain important finally manual interpretation by professionals is still re-
research results. However, they are lengthy and rich in quired. According to a survey of examiners at the European
technical terms, so analysis requires a lot of manpower, and Patent Office (EPO), 84.7% of examiners believe that CPC is
there is an urgent need for automatic tools to assist patent very important for patent searches. Although 70% of ex-
engineers or decision makers in patent analysis. The im- aminers believe that AI and ML technologies can provide
portance of patent mining is thus seen. Patent-mining valuable support in the future, about 45% of examiners still
technology includes text segmentation, abstract extraction, believe that patent searches fundamentally rely on human
feature selection, term association, cluster generation, topic efforts. And 52% of examiners do not think that a fully
identification, and information mapping [28]. In addition to automated patent search can be done before 2035 [38].
the extensive use of LDA topic modeling methods in on- In the practice of the industry, most of the patents
tology construction, it is also very popular in patent mining. collected are read by the researcher one by one and classified
In the patent analysis application of drones, through according to the technical field and effect of their profes-
LDA, the three most active technology development themes sional human judgment. The manual classification method
such as communication technology, power supply, and consumes a lot of time, and it is difficult to obtain a
navigation system are found [29]. Based on LDA, Korobkin comprehensive review through the interpretation of a large
et al. [30] proposed a new patent-mining method, which number of patent documents. Many recent studies have tried
includes statistical and semantic analysis of patent docu- to find a more efficient way to construct TFM. Yang and Ren
ments, machine translation of patent applications, and [39] proposed a semiautomatic TFM construction method
calculation of semantic similarity between patents and ap- by extracting technical words and computer-aided algo-
plications. In the aspect of term association, Hu et al. [31] rithms to reduce labor costs and time. Ki and Kim [40]
utilized a skip-gram-based model to extract key terms from proposed a programmatic automation method based on
patents and compared the proposed approach with the TF- NLP technology to quickly construct an Information Re-
IDF method. In terms of cluster generation, k-means is still lation Matrix (IRM), which describes relationships among
powerful. Shanie et al. [32] used the k-means method to technical information in the patent and is similar to TFM.
cluster patent documents related to green tea, in which the Trappey et al. [41] used the resultant patent text and data
Complexity 5

mining technology to create ontology-based TFM for patent postprocessing. In addition, this research also performed
analysis of additive manufacturing in the dental industry. patent management map analysis and compared the results
The abundant literature shows that ontology, text mining, with text-mining to explore emerging technologies and
NLP, topic modeling, and TFM technology can be regarded verify the ideas and conclusions put forward in this research.
as the main procedures for patent analysis today.
3. Patent-Based Ontology Construction
2.6. Comparison. Table 1 shows the comparison of the 17 Figure 1 illustrates the ontology construction process, in-
related studies of technology, especially patent-mining cluding four levels and two aspects.
techniques from its research purposes, tasks for pre- The four levels are patent retrieval, patent clustering and
processing, processing, and postprocessing [16]. The second target domain selection, topic modeling, and keyword
column of Table 1 lists the tasks in each part, and the third generation. The two aspects are research process and on-
column lists the more specific methods used. Each successive tology construction. At level 1, some key terms about natural
column corresponds to each article, of which part, task, and language-enabled chatbot are figured out, and the smart
method used are listed. search on DI is used to do the patent retrieval. Then, the
In the preprocessing part, the use of natural language most related 50 patents are quickly glanced to check if they
processing for text preprocessing is mentioned in most match the subject of this study. If not, the search query is
articles, and the corresponding algorithms, tools, or kits are adjusted and do the retrieval again until the records are
quite mature. Although some articles did not specifically much in line with the subject. At level 2, DWPI title, DWPI
mention this part, it is believed that this part, as a relatively abstract, and independent claims are used to do the k-means
mature part, should have been implemented. Two main clustering, and silhouette score is used to evaluate the
tasks, key term extraction and patent management map, are propriate number of clusters. After clustering, normalized
included in the processing part. The TF-IDF method is TF-IDF (NTF-IDF) is used to identify the key words and key
widely used in the key term extraction task and can almost be phrases. Again, we will check if the key words match the
regarded as a standard configuration. Skip-gram is an im- subject. If not, go back to level 1 and adjust the search query.
portant method to study the contextual relationship, and it is Repeat the process until ideal target domains are found. At
often used in the research that uses the contextual rela- level 3, topics for domain are found in 2 different ways. The
tionship as the vectorization method. Patent management LDA model is used in domain of NLP, model, and system,
map, or patent map analysis, is a statistically-based data while manual induction is used in domain of applied sce-
analysis method that has been widely used, with a database narios. In order to discover deeper topics or concepts at this
and business intelligence tools to visualize patent portfolios. level, each domain resets patent search conditions for ap-
Patent management map only involves data sorting and plying the LDA method. After each execution, it is deter-
presentation, which does not conform to the current general mined whether the subject of each domain is clearly
definition of text-mining. Therefore, it is hardly mentioned identified according to the results. If not, the patent search
in the research of patent analysis by text-mining in recent conditions must be adjusted again. The topics of each do-
years. Among them, only the patent classification code will main are determined in this iterative process. Finally, by
be referenced as a benchmark to verify whether the results of sorting out the key words and key phrases from level 2 and
the text-mining-based approach are valid and consistent. level 3, the construction of level 4 can be completed.
The postprocessing part contains two parts: text-mining-
based approach and visualized approach. The main methods
of the former are clustering, topic modeling, and classifi- 3.1. Patent Retrieval. Smart search on DI provides a se-
cation; the latter is mostly based on the expression of node- mantic search tool, which offers a quick path to capture
relation graph. Although TFM is less common, it is still one related patents from simple search terms. The powerful
of the good visualization tools for exploring emerging algorithm behind replicates the strategies used by expert
technologies. searchers to provide a manageable result set that matches
The main purpose of these studies is focused on clas- users’ intent. By using smart search, it is not necessary to list
sification, ontology construction, and finding emerging all probable related terms before searching. Instead, the
technologies. Classification is very basic, and the patent data records discovered are always related to the technology
itself already have classification codes, such as IPC or CPC. described by the input terms but may not be exactly con-
Researchers who use classification methods in post- tained. Smart search automatically sorts the result set
processing parts have a clear aim at classification. Ontology according to the relevance score to show the content that
construction aims to clarify the technical details and scope of best matches the search term.
a specific field, and clustering and topic modeling methods In order to obtain a well-constructed ontology, the main
can achieve this goal well. Both classification and ontology purpose is to find as wide a range of technologies as possible
construction only obtain and analyze existing data, but in from the field, and not to focus on specific technologies that
order to explore emerging technologies, it is necessary to will lead to a small number of emerging technologies that
find rules or discover changes in trends from the data. cannot be found. Smart search has the advantage of intel-
The framework proposed in this study completely in- ligence, but the limit of 1,000 records corresponds to about
cludes the three parts of preprocessing, processing, and 450 to 550 DWPI families on average, which is not much in
 6

Table 1: Comparison for studies related to patent and technology mining techniques.
Proposed
Part Task Method [18] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [37] [39] [40] [41]
framework
Preprocessing Data processing Text preprocessing ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
TF/TF-IDF ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Key term extraction
Skip-gram ✓ ✓ ✓ ✓
Processing Yearly trend ✓ ✓ ✓
Patent management map Assignee ✓ ✓
CPC/IPC ✓ ✓ ✓ ✓
Clustering ✓ ✓ ✓ ✓ ✓ ✓
Text-mining-based
Topic modeling ✓ ✓ ✓ ✓ ✓ ✓
approach
Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Postprocessing Semantic tree
✓
analysis
Visualized approach
Node-relation graph ✓ ✓ ✓ ✓
TFM ✓ ✓ ✓ ✓ ✓
Prior art patent search ✓
Classification ✓ ✓ ✓ ✓ ✓ ✓
Purpose Ontology construction ✓ ✓ ✓ ✓ ✓
Finding emerging
✓ ✓ ✓ ✓ ✓ ✓
technologies
Complexity
Complexity 7

Research process Ontology construction

Level 1: patent retrieval
Refine search query
Set search query
False
Match
Patent retrieval
subject
True

Level 2: patent clustering and target domain selection

Patent columns for False

Match
clustering subject

True

Decide number of
clusters

Clusters
Clustering Domains
(key words/phrases)

Level 3: topic modeling

(For each domain)
Refine search
query Set domain search query
and do patent retrieval

Technologies, functions Domain Applied

scenarios
type
LDA topic modeling

Manual induction
Topic-word
generation

False
Topics True
Topics
found

Level 4: key words/phrases

Key words/phrases

Figure 1: The ontology construction process flow.

terms of the number of patents related to NLP chatbot. The several rounds of trials this study finally selected 508 DWPI
results of patent search will be used as the data source for families detected by smart search as the results of level 1, and
clustering task at level 2. To use more patents for clustering, its search is shown in Table 2.
traditional patent search is also tried, which directly search
patents form the original term user lists. Although by tra-
ditional search more patents can be found, if there are 3.2. Patent Clustering and Target Domain Selection. At level
emerging technologies or applications that are not widely 2, the patent obtained from the previous level is clustered
discussed or even undetected, they will not be found. After and some target domains are discovered from the results.
8 Complexity

Table 2: Search query for clustering.

Search type DI query Result
Smart SSTO � (“natural language processing” “natural language understanding” “NLP” “NLU” “chatbot”
508 DWPI families
search “VIRTUAL ASSISTANT” “INTELLIGENT ASSISTANT” “automated conversational interface”)

The process begins at extracting the words in the patent scenarios” domain is in clusters 1 and 2. Some clusters are
document and using NTF-IDF to do vectorization, so that related to multiple domains at the same time. Since the
numeric vectors are obtained and can be applied to perform purpose of cluster analysis is to find out target domains, it is
the k-means clustering. After that, the top words and n-gram not so important whether each group must be clearly
top phrases of each cluster can be counted, from which target assigned to only one certain domain.
domains are selected. This research takes natural language-enabled chatbot as
the subject. A large number of words related to NLP appear
in large numbers in each cluster, which is not helpful to find
3.2.1. Patent Columns for Clustering. This study chooses out the domain, such as “NLP,” “natural language,” and
DWPI title, DWPI abstract, and independent claims as the “processing.” In addition, many chatbot-related words are
source attributes for clustering. Patent documents may come very versatile, which also increase the difficulty of domain
from different countries, written in different languages, and exploration, such as “processor,” “request,” “input,” and
cover a large number of attributes. Patent is to protect the “module.” The above vocabularies are skipped during the
inventor’s smart finance or as a consideration for the en- domain selection. One step in the preprocessing of patent
terprise’s knowledge layout. Contrary to academic articles, documents before clustering is to vectorize the patent
patents are not written for users to understand easily, and documents. Although those skipped terms in Table 4 could
some information may even be deliberately hidden in the be set as stop words in the preprocessing stage, the reason for
title, which is not conducive to patent mining. The DWPI not skipping them is to avoid affecting the integrity of some
title and DWPI abstract, provided by the DI database, just phrases. Take “recognition” as an example. “Recognition” is
solve the above problems. DI employs discipline-profes- also included in “intent recognition,” “named entity rec-
sional editors with scientific and engineering backgrounds to ognition,” “speech recognition,” and “image recognition.”
manually read all patents one by one and rewrite the title and While setting “recognition” as a stop word, the above related
abstract with easy-to-understand text, which are DWPI title phrases will not be found. However, failing to remove
and DWPI abstract, respectively. They remove the legal “recognition” has caused it to appear repeatedly in each
jargon, use American spelling, and intellectually choose cluster and does not have domain recognition.
drawing instead of just choosing the ones on the front page. “NLP” domain contains cognition, named entity recog-
In addition, many studies have shown that the value of nition (NER), linguistics (which include syntactic, semantic,
patents is greatly affected by the number of independent and morphology), natural language understanding (NLU),
claims, which are also included as the source attribute of the response, and speech recognition. Nine clusters, cluster 3, 4, 5,
cluster. 6, 8, 9, 11, 12, and 13, are distributed in NLP domain. For
cluster 3, two subdomains, cognition and response, are in-
3.2.2. Clustering. After retrieving and vectorizing patent volved. For cognition subdomain, representative patent
documents, k-means can be performed to show the clus- US9361884B2 (assignee: Nuance Communication Inc.) pro-
tering distribution phenomenon in the vector space. The posed a human-machine dialogue system, incorporating with
appropriate number of clusters can be obtained by calcu- an NLU engine and a dialogue manager for providing NLP
lating the silhouette score: the goal is to maximize the application to identify and resolve anaphora. For response
distance between clusters and minimize the distance within subdomain, patent US10417266B2 (assignee: Apple Inc.)
clusters. In this study, 13 clusters are clustered from 508 proposed systems and processes for operating an intelligent
patents, and the top 10 words and 2-gram phrases in each automated assistant to provide a set of predicted responses.
cluster are extracted through NTF-IDF (see Table 3). Cluster 4 and 5 focus on linguistics. Patent US20200327284A1
(assignee: ServiceNow Inc.) in cluster 4 proposed an agent
automation system, which has processor that is configured to
3.2.3. Domain Selection. The top 10 words and 2-gram assign respective word vector to nodes and encodes semantic
phrases of 13 clusters, with a total of 260 terms, of which meaning of word or phrase represented by nodes. The system
technical details are examined individually, are classified as generates an annotated utterance tree by using a combination
13 subdomains, which are combined to form the 4 domains, of rule-based and ML-based components, wherein an an-
that is, NLP, model, system, and applied scenarios (see notated utterance tree represents a syntactic structure of the
Table 4). The subdomains can display related topics and utterance, and nodes of the annotated utterance tree include
assist topic selection when performing topic modeling in word vectors that represent semantic meanings. The anno-
level 3. “NLP” domain is distributed in clusters 3, 4, 5, 6, 8, 9, tated utterance tree is used as a basis for intent or entity
11, 12, and 13; “model” domain is concentrated in cluster 6; extraction. Patent EP3111338A1 in cluster 5 also used au-
“system” domain has clusters 7, 9, and 10, and “applied tomated text annotation for the construction of NLU
Complexity 9

Table 3: Top 10 words and 2-gram phrases in each cluster.

Cluster Size Top 10 words and 2-gram phrases
Assistant, automate, user, input, language, natural, client, human, computer, processor
1 20 Automate assistant, natural language, automated assistant, virtual assistant, assistant client, automate summarization,
human computer, computer dialogue, input corpus, telephone call
Engine, language, natural, user, medical, code, processing, billing, generate, clinical
2 27 Natural language, medical billing, billing code, language processing, language understand, patient encounter, clinical
patient, free text, question answer, processing engine
User, request, response, language, natural, processing, action, query, input, generate
3 45 Natural language, user request, action structure, language processing, language request, response user, speech input,
dynamic training, computer readable, request text
Word, language, natural, phrase, computer, processing, plurality, sentence, processor, clause
4 31 Natural language, target word, language processing, word clause, word phrase, neural network, input question, numeric
code, program instruction, user interface
Text, language, natural, processing, user, processor, process, semantic, information, computer
5 46 Natural language, language processing, language text, language understand, text interest, input text, semantic segment,
information processing, touch operation, text natural
Plurality, language, entity, natural, generate, computer, associate, processing, document, name
6 52 Natural language, name entity, language processing, computer readable, language input, cluster classification, reduced
aggregation, flow diagram, machine learn, neural network
User, interface, language, NLP, natural, display, input, query, information, computer
7 64 Natural language, user interface, language processing, user input, user query, graphical user, processing NLP, voice
apparatus, real time, computer readable
speech, user, input, recognition, processing, language, determine, computer, audio, natural
8 30 Natural language, speech recognition, speech processing, language processing, input audio, speech input, computer
implement, user profile, language understand, automatic speech
Input, user, intent, language, natural, determine, processor, generate, NLU, computer
9 52 Natural language, language input, user input, user intent, voice input, language understand, computer readable, input
determine, transitory computer, user interface
Communication, user, language, natural, input, interface, voice, computer, processor, call
10 35 Natural language, voice input, language processing, communication interface, input communication, text communication,
voice communication, user input, phone call, communication channel
Application, user, language, natural, NLU, input, computer, associate, plurality, processor
11 27 Natural language, online application, speech word, user online, language processing, language input, part speech, language
understand, dimensional vector, structured natural
Information, language, module, natural, user, entity, service, input, generate, obtain
12 35 Natural language, touch screen, language understanding, language understand, language processing, object hovering,
understanding module, component process, target conversation, question answer
Language, natural, program, processor, structure, computer, instruction, user, analysis, expression
13 44 Natural language, program instruction, language expression, frame structure, language processing, computer readable,
language understand, semantic structure, language story, computer program

grammars. Patent US10789426B2 in cluster 5 described a contextual voice user interface. Patents US9245525B2,
device for processing natural language text with the context- US9741347B2, and US10049676B2 describe an interactive
specific linguistic model. Patent US10304444B2 (assignee: response system mixes HSR subsystems with ASR subsys-
Amazon Tech Inc.) applies NLU to the music field, which uses tems to facilitate overall capability of user interfaces. Patents
a hierarchical organization of intents and entity types, and US9245525B2, US9741347B2, and US10049676B2 describe
trained models associated with those hierarchies, so that an interactive response system mixes HSR subsystems with
commands and entity types may be determined for incoming ASR subsystems to facilitate overall capability of user
text queries without necessarily determining a domain for the interfaces.
incoming text. Although cluster 6 is mainly concentrated in Cluster 9 mentions about NLU, in which patent
the “model” domain, there are also many terms related to US9761225B2 (assignee: Nuance Communications Inc.) is
“named entity.” A representative patent US10755046B1 representative. In US9761225B2, a method for identifying
(assignee: Narrative Science) describes an NLP system for and resolving anaphora in multimodal conversational dia-
conversational inferencing with four-step parsing process. logue application for smartphone is proposed, in which
Cluster 8 focuses on speech recognition. Patents multiple NLU interpretation selection models may be
US10446147B1 and US20200118564A1 (assignee: Amazon generated. The NLU interpretation selection models may
Tech Inc.) describe a speech recognition system to provide a include a generic model and one or more specialized NLU
10 Complexity

Table 4: Domains, subdomains, and terms.

Domain Subdomain Cluster Key words/phrases
Action, expression, action structure, frame structure, semantic structure,
Cognition 3, 13
automate summarization
Named entity recognition 6 Entity, name, name entity
Part speech, word, phrase, sentence, semantic, plurality, document, clause,
Linguistics (syntactic semantic,
4, 5 expression, target word, word clause, semantic segment, semantic structure,
morphology)
NLP language expression
NLU, user intent, user request, language understand, language understanding,
NLU 9, 11, 12
understanding module
Response 3, 12 Generate, question answer, response user
Speech, voice, audio, speech recognition, voice input, speech input, speech
Speech recognition 8
processing, input audio, speech word, automatic speech
Engine, machine learn, program, neural network, cluster classification,
Model Model 6
processing engine, dynamic training, dimensional vector, reduced aggregation
Human, display, call, user interface, user input, touch operation, online
User interface 7, 11 application, phone call, telephone call, user online, graphical user, real time,
human computer, touch screen, object hovering
System Computer, voice apparatus, computer readable, voice input, transitory
Medium 9
computer,
Communication, communication channel, communication interface, text
channel, communication 10
communication, voice communication
Service, assistant, automate, client, automated assistant, virtual assistant,
Personal 1
Applied assistant client
scenarios Medical, billing, clinical, medical billing, billing code, patient encounter, clinical
Medical 2
patient
User, NLP, natural, language, interface, computer, analysis, query, recognition,
process, processor, processing, structure, program, code, application, request,
response, input, obtain, associate, service, instruction natural language,
language processing, processing NLP, language input, input text, text interest,
flow diagram, language text, determine, computer program, language input,
Skip Skip All
information, input determine, input corpus, computer dialogue, free text, user
query, language input, language request, input question, text natural, module,
user input, user profile, component process, language story, structured natural,
program instruction, frame structure, information processing, numeric code,
computer implement

interpretation selection models, and each of which may be integration of technologies and functions are in has come an
specific to a particular set of NLU interpretation type. Se- important basis for judging technicality. If AI is only used to
mantic reranking mechanism is applied in this method. analyze business data, and technical problems are not solved,
Cluster 11 also mentions about NLU capability and focuses it is likely to be regarded as having no technical ideas, and it
more on the follow-up actions, which are more related to is difficult to overcome the nonpatent reasons by applying
“system” domain. Cluster 12 focuses on knowledge ex- for repetition or amendment [42]. Algorithm-related patents
traction in NLU. The representative patent is must be combined with hardware-related terms as the
US10762113B2, which uses conversational knowledge carrier of the algorithm. This also explains the reason why
graphs in virtual assistants to process natural language input, cluster 9 contains a large number of “nontransitory com-
which involves receiving natural language queries from users puter readable device” vocabulary. The representative pat-
at the virtual assistant’s NLU system. Cluster 13 also belongs ents in cluster 6 are US10748526B2, US10747958B2, and
to cognition subdomain. Patents US9965461B2, US10733375B2.
US9594745B2, US9569425B2, and US20140249801A1 in “System” domain contains user interface, medium, and
cluster 13 (assignee: The Software Shop Inc.) describe the communication or channel subdomains, in which four
method for improving efficiency of syntactic and semantic clusters, cluster 7, 9, 10, and 11, are distributed.
analysis. As for “applied scenarios,” concentrated in cluster 1 and
“Model” domain, concentrated in cluster 6, has no 2, terms such as “virtual assistant,” “medical,” and “billing”
subdomain, and the number of key words is relatively low. are found. In cluster 1, three patents assigned to Google LLC
The possible reason is that since neural networks are mainly are representative for virtual assistants in “personal” sub-
mathematical algorithms and computers are only the car- domain. Patent US20200320136A1 proposes a method for
riers of mathematical operations, they cannot contribute to using distributed state machines for human-to-computer
the technology themselves. In this case, what field the close dialogues with automated assistants to protect private data.
Complexity 11

Patent US20200050788A1 describes a system for assembling storing situation information collected from a vehicle. A
responses from remote automated assistants. Patent dialogue management module obtains a factor value of
KR2020131299A proposes a method for generating Internet action factor used to perform an action corresponding to a
of things-based notification by automated assistant client of dangerous situation when an input processor obtains an
client device. In cluster 2, three patents assigned to Nuance action corresponding to the starting situation from the
Communication Inc. are representative for medical billing storage unit. An input processor generates a dialogue to
and coding in “medical” subdomain. Medical billing and perform the action corresponding to the dangerous situation
coding are two closely related aspects of the modern health by using the factor value of the acquired action factor while
care industry. Both practices are involved in the immensely obtaining the action corresponding to the dangerous situ-
important reimbursement cycle, which ensures that health ation and generates a conversation message. A result pro-
care providers are paid for the services they perform [43]. cessor generates a conversation response corresponding to a
Patent US20170323060A1 describes a system with a delivered starter message. Patent US10223934B2 proposes a
graphical user interface (GUI) and an NLU engine to au- method for monitoring and analyzing language environ-
tomatically derive one or more engine-suggested medical ment, vocalization, and development of key child, which
billing codes. Patent US10319004B2 proposes techniques to provides metrics associated with key child’s language en-
deal with the overlapping codes derived by the NLU engine, vironment and development in a relatively quick and cost-
and patent US10754925B2 proposes a method for training effective manner. The proposed method is used to promote
NLU engine, involves providing training data in form of improvement of the language environment and key child’s
free-form text, corrections, and finalized sequence of language development and to track development of the
medical billing codes. child’s language skills. Key child’s language environment
Three domains were found from the clustering results. It and language development are monitored without placing
is particularly important to emphasize that the composition artificial limitations on the key child’s activities or requiring
of natural language-enabled chatbot mostly relies on the third party observer.
three domains, NLP, model, and system. Since most of the Here are some patents in topic of engineering. Patent
related patents contain these three parts at the same time, it JP06792132B2 defines an information-processing apparatus,
is difficult to determine the exact belonging domain for each which is used in the manipulator control system and NLP
patent and also meaningless. system and can be performed with high versatility. The
information-processing apparatus has processing module
groups, and each of which is equipped with several pro-
3.3. Topic Modeling. According to the ontology construction
cessing modules with specific processing capabilities. These
process (see Figure 1), search query, corresponding result,
processing modules have a neural network with a hierar-
and topics founded in each domain are illustrated in Table 5.
chical structure. The information is processed by sending
For domain NLP and system, DI smart search is applied,
and receiving the information signal of the processing
while CTB (claim/title/abstract) strategy is applied for do-
module in several interhierarchical structures. Patent
main model and applied scenarios. Table 6 illustrates the
CN111267097A proposes a natural language-based assisted
keywords of each topic.
programming method for industrial robots, involves parsing
language instructions, matching parsing result, and com-
3.4. Applied Scenario Topic Modeling. This research hopes to bining coordinates output to generate final robot auxiliary
find the application field of NLP chatbot, but a lot of experts code. The multiattention mechanism model adopted by the
are describing natural speech-related technologies or the method improves the recognition accuracy and solves the
system framework of conversation management, which are problem that the current method cannot accurately recog-
not discussed in this section. This research mainly divides nize objects in an industrial environment. Modular pro-
the application scenarios into engineering applications and gramming technology solution simplifies engineers
e-commerce applications. It can be found from the patent programming complexity and effectively improves devel-
search results that natural language-enabled chatbot is opment efficiency. Patent US10843080B2 describes a system
widely used in the field of e-commerce, while the application for facilitating automated program synthesis from natural
on the engineering side is difficult to find. 44 patents are language. The system allows a user to be more comfortable
reviewed manually and classified to certain topic or scenario. and familiar with grammatical requirements for forming a
These patents with respect to the applied scenario are listed proper sentence in native language as opposed to memo-
in Table 7. rizing rules or required constructs for a potentially com-
Here are some patents in topic of e-commerce. Patent plicated programming language. The system employs fuzzy
US20170323060A1 describes a system for facilitating auto- grammar matching to reduce complexity, while slightly
mated natural language understanding for medical docu- trading off complexity for accuracy. The system allows the
mentation of patient, which has processor for presenting set user or developer to examine to express an idea in a different
of medical billing codes for user review in graphical user manner to better reflect user an original intent. Patent
interface (GUI) before finalizing coding of encounter. Patent DE102018212503A1 defines communication and control
KR2020000621A describes a conversation system for systems, which has control devices for operating machine
grasping user attention during various situations in a vehicle based on software communication chatbot, for filling bev-
by using a mobile device. The system has a storage unit for erage in bottling plants. The chatbot recognizes a voice input
12 Complexity

Table 5: Topics in each domain.

Query
Domain/method
Search type Input patent size Topics
SSTO � (“natural language processing” “linguistics” “natural language generation” “natural language
NLP/LDA understanding” “speech recognition”)
Smart 570 Linguistics, conversation, speech recognition, knowledge
CTB � ((chatbot) or (automated adj conversation∗ adj interface∗ ) or (chat∗ ADJ system∗ ) or (natural adj
language∗ ) or (nlp∗ )) AND CTB � (((deep ADJ learning) or (machine ADJ learning) or (neural ADJ network)))
Model/LDA
AND DP> � (20200101) AND DP< � (20201130)
CTB 2,535 Features, voice device, question answer, classification, graph, automatic service
SSTO�(“natural language processing” “natural language understanding” “NLP” “NLU” “chatbot” “automated
conversational interface”) AND SSTO � (“user interface” “medium” “communication” “channel” “immersive
System/LDA
technology” “computer vision”)
Smart 534 User interface, dialogue management, infrastructure
CTB � (((chatbot ) or (conversation ) or (diolog∗ adj system∗ ))) AND CTB � (((natural adj language∗ ) or
∗ ∗
Applied scenarios/
(nlp∗ ))) NOT DC � ((T))
manual
CTB 31 Engineering, e-commerce

Table 6: Keywords for each topic.

Domain Topic Keywords
Linguistics Personality, AI, discourse, syntactic
Conversation NLU, semantic, NLG, intent
NLP
Speech recognition Audio signal, processor, channel
Knowledge Entity, ontology, semantic, cognitive identification
Features semantic, vector representation, image recognition
Voice device Storage, server, control module
Question answer Pair, retrieval, RNN
MODEL
Classification Segmentation, convolutional, encoder
Graph Entity, ontology, intent
Automatic service Call, recommendation
User interface Portable, network, wireless, digital
System Dialogue management NLG, processor, ML
Infrastructure Channel, cloud, communication

Table 7: The manual induction result by applied scenario.

Topics Scenario Publication number
US20170323060A1 WO2020061562A1 US10679345B2
Medical US20200185102A1 JP2020518047A US20200027535A1
CN111612752A US10754925B2 US10319004B2
Health CN109591024A US10748644B2
KR2020000621A WO2020069517A3 US10752212B2
Driver assistant CN111145731A US20200216089A1 US10543931B2
US10573299B2 US20200135183A1 EP3606797A2
E-commerce Exercise KR2173553B1 US20200114207A1
Education US10223934B2
Emotion US10579742B1 CN111312394A
Smart home IN202041050057A CN110654738A
Customer service CN108282587B IN201821029643A CN111902878A
US10748526B2 US10747958B2 US10733375B2
Smart assistant
EP3753017A1
Entertainment EP3566399A4
CN111645073A CN111267097A JP2020526402A
Robot
JP06792132B2 US20200306958A1
Engineering Programming US10843080B2
Manufacturing CN107632845B DE102018212503A1
Quality control WO2020181365A1
Complexity 13

and a text input by an operator to output or display in- past decade. However, starting in 2016, the annual growth
formation about an operating state of the machine. The rate has increased sharply until it reaches a peak of 105% in
systems realize production conversion of energy in an au- 2019, and it then falls back to 66% in 2020. Whether the
tomatic manner and order completion in a rapid manner decrease in the number of 2020 is related to the impact of
and improve media efficiency and scheduling efficiency. COVID-19 is unknowable, but this may be a signal that
Patent WO2020181365A1 proposes an apparatus for 360- implies that the technology related to natural language-
degree assistance for quality control system scanner with enabled chatbot may have gradually matured.
mixed reality (MR) and ML technology. The apparatus has However, a single reduction in quantity cannot lead to
an optical sensor, a display, and a processor to receive di- any conclusions unless supported by more other data or
agnostic information from a server related to a field device in evidence. IPC is a standard taxonomy developed and ad-
an industrial process control and automation system. The ministered by WIPO for classifying patents and patent
processor identifies an issue of the field device based on the applications, which covers all areas of technology and is
diagnostic information, detects, using the optical sensor, the currently used by the industrial property offices around the
field device corresponding to the identified issue, and guides, world. From the annual number of patents with IPC
using the display, a user to a location and a scanner portion analysis, to 2018, all The IPC classifications have been
of the field device that is related to the issue. The processor covered. In other words, among the 8,099 patents in 2019
provides, using the display, necessary steps or actions to and 2020 that accounted for 62% of the number in the past
resolve the issue, and connects, using a cloud server, a user to decade, no new technology has been produced.
get modules of installation, commissioning, AMC, and Top 6 4-character IPCs, with a number of patents that
training for a QCS as per the selected person. greater than 1,000, are G06F (electric digital data process-
ing), G06N (computer systems based on specific compu-
tational models), G06Q (data processing systems or
3.5. Ontology. In this section, the ontology map of NLP methods), G10L (speech analysis or synthesis; speech rec-
chatbot is drawn based on the previous outputs. A four-level ognition; speech or voice processing; speech or audio coding
ontology includes subject, domains, topics, and key phrases or decoding), H04L (transmission of digital information),
in a top-to-bottom sequence. Under the subject of NLP and G06K (recognition of data), each in which has a number
chatbot, the domains are NLP, model, system, and applied of 8,870, 3,144, 2,413, 2,176, 1,364, and 1,258 patents, re-
scenarios. The third level has the topics under each domain. spectively (see Figure 3). It should be noted that the total
For NLP domain, there are speech recognition, linguistics, proportion can exceed 100%; that is, the summation of these
conversation, and knowledge. For domain of model, topics number can be greater than 12,840, because a patent can be
are feature, graph, voice device, question answering, clas- classified as multiple IPC codes.
sification, and automatic service. For domain of system, the G06F’s patents accounted for 8,870 of 12,480 patents.
topics are infrastructure, dialogue management, and user Therefore, the complete IPC classification in G06F was
interface. For applied scenarios, e-commerce and engi- further explored. Among the top 10 IPCs listed (see Fig-
neering are the two main topics. The fourth level has the key ure 4), 2,295 patents are classified in G06F 17/27 (for au-
phrases under each topic. It is noticed that some key terms tomatic analysis, parsing, orthographic correction, etc.). The
are shared by multiple topics. The ontology map of NLP second largest class is G06F 17/30 (for information retrieval
chatbot is shown in Figure 2. and database structure). It is worth noting that the 3rd and
4th classifications (G06N 3/04 and G06N 3/08) represent the
4. Patent Macro Trend Analysis interconnection topology architecture and learning method,
respectively. G06F focuses on data processing procedures,
Related patents are searched by entering keywords related to while G06N emphasizes system structure. G06F and G06N
NLP and chatbots on the DI database, and patent man- domain classifications represent the key technologies for
agement map analysis is conducted (see Table 8). From 2011 implementing the main modules of complex natural lan-
to 2020, totally 21,834 individual records or 12,840 DWPI guage-enabled chatbot systems. In addition, G10L 15/22,
families are published. Patent family refers to the collection ranked 9th, is about programs used in speech recognition for
of patents applied for in different patent offices for the same human-machine dialogue.
invention. DWPI has a stricter definition. Each patent in the In addition to statistics on the number of patents, the
same DWPI patent family must have exactly the same fluctuations in the number in recent years are also worthy of
priority as other patents in the family. The analysis of this attention. Based on the annual growth rate of all patents,
section is mainly based on DWPI families. The following when the growth rate of an IPC is higher than average, it
term “patents” refers to “DWPI families” unless otherwise represents greater momentum; conversely, when the growth
specified. rate of an IPC is lower than average, it may imply that the
Since 2017, 10,480 patents have been published, ac- technology has entered the mature stage early. The four 4-
counting for 82% of the total 12,840 patents in the past character IPCs with the largest number were selected for this
decade. Furthermore, since 2019, 8,099 patents account for analysis (see Figure 5).
62%. From the perspective of the annual growth rate of the G06F has an overwhelming 69% of total patents, but its
number of patents, the number was a high 44% in 2014, but annual growth rate is much inferior to the average annual
returned to 6% in 2015, which is the lowest number in the growth rate. In 2014, the total number of patents related to
14 Complexity

Syntactic Portable
Audio signal Processor Channel Cloud Communication ML
Discourse Network
Speech
AI recognition Dialog User Wireless
Infrastructure
management interface
Personality Linguistics Digital

NLG System Medical

NLU Conversation Driver

NLP Applied
NLP Ecommerce assistant
chatbot scenarios
Intent Customer
service
Engineering
Semantic Model
Knowledge Robot

Cognitive
Voice Question Automatic Programming
identification Feature Graph Classification
device answering service
Ontology Manufacturing
Storage Pair Encoder
Entity
Server RNN CNN Call

Vector Image Control module Retrieval Segmentation Recommendation

representation recognition

Figure 2: NLP chatbot ontology.

Table 8: Search query for patent management analysis.

Search type DI query Result
CTB�((chatbot) or (automated adj conversation∗ adj interface∗ ) or (chat∗ ADJ system∗ ) or (natural
adj language∗ ) or (nlp∗ )) AND CTB�(((ontology) or (named ADJ entity ADJ recognition) or (deep
ADJ learning) or (machine ADJ learning) or (neural ADJ network) or (speech ADJ recognition) or
(feature∗ ) or (lstm) or (gated adj recurrent adj unit) or (transformer) or (BERT) or (GPT∗ ) or
Claim/title/ (rectifier) or (RELU) or (“speech%” ADJ “generat%”) or (cloud ADJ computing) or (voice ADJ 12,840 DWPI
abstract activity ADJ detection) or (voice ADJ over ADJ Internet ADJ protocal) or (bandwidth) or (human families
ADJ computer ADJ interaction) or (VUI) or (GUI) or (user ADJ interface) or (immersive ADJ
technolog∗ ) or (virtual ADJ reality) or (augmented ADJ reality) or (mixed ADJ reality) or (force ADJ
touch) or (3D ADJ touch) or (robotic ADJ process ADJ automation) or (communication ADJ
system))) AND DP>�(20110101) AND DP<�(20201231);

Top 10 IPCs (4 characters)

10000 8870
8000
6000
4000 3144
2413 2176
2000 1364 1258 618 354 296 283
0
G06F G06N G06Q G10L H04L G06K G16H H04M G06T H04N
Figure 3: Top 10 IPCs (4 characters).

natural language-enabled chatbot rose sharply by 44.37%. period of rapid growth in the number of patents, the growth
The growth rate of G06F in that year was only 41.40%, which rate of G06F has not been outstanding. Even when the
was slightly lower than the average. Since 2016, during the average growth rate reached a peak of 104.49% in 2019, G06F
Complexity 15

Top 10 IPCs in G06F

2500 2295
2000 1667
1500
900 873
1000 723 720 697 664 661 655
500
0
G06F G06F G06N G06N G06F G06F G06F G06F G10L G06F
17/27 17/30 3/04 3/08 40/30 17/28 16/35 16/332 15/22 16/33
Figure 4: Top 10 IPCs in G06F.

100.00 89.49%
74.29%
73.74%
50.71% 57.28% 52.84%
43.86%
50.00 23.98% 26.14% 21.59%
3.79% 12.97% 23.36% 20.51%
5.56% 9.47% 9.96% 4.35% 9.05% 6.89%
1.80% 1.64% 2.09% 4.02%
% 0.00
–1.67% –2.98% –3.33% –7.49%
–13.96% –8.42%
–20.30% –20.86% –14.92% –19.62%
–24.13%
–50.00 –34.85%

–100.00
2012 2013 2014 2015 2016 2017 2018 2019 2020
G06F G06Q
G06N G10L
Figure 5: Annual patent growth rate under top IPCs.

was 14.92% less than the average. By contrast, the annual increase rapidly at the same time. In addition to the two
growth rate of G06N is amazing. In 2014, it was 43.86% categories of G10L 15/18 and G10L 15/22 related to speech
higher than the average, and from 2016 to 2020, the annual recognition in 2019, both Google and Samsung have more
growth rate was 73.74%, 26.14%, 89.49%, 52.84%, and patents appearing in G06F 3/16, which focuses on the
74.29% higher than the average, respectively. G06Q and conversion between speech and digital information. On the
G10L fluctuate up and down in average annual growth rates other hand, Google and Baidu applied for many patents on
and have not yet shown a clear trend. G06N 3/08, which are the computer system based on
In general, the average annual growth rate began to learning methods. In addition, Baidu also has a large number
slow down after reaching a peak in 2019 after rapid of patents on G06F 40/30 for semantic analysis. Google and
growth, no new IPC appeared after 2018, and all of which Baidu are both Internet service companies that started as
indicate that the development of natural language- search engines, and Google and Samsung are also close
enabled chatbot has entered a mature stage. It is worth partners in the android camp. The highly increasing number
noting that the patents related to only reading G06N are of patents assigned to these three companies, which are quite
still growing rapidly. close to the end user, might imply the maturity stage and
Assignee analysis helps to find the main players in the mass application in this technology field. From the IPC
market, which are all technology giants from the results. The distribution of Apple Inc.’s patents in 2019 and 2020, it can
number one IBM has 1,358 patents, which is more than the be seen that its patents are highly concentrated on speech
total number from the second to the tenth. The well-known recognition-related G10L 15/18, G10L 15/22, and G06F 3/
technology giants Apple Inc and Facebook Inc are ranked 16, which are similar to Google. Google and Apple coin-
16th and 17th, respectively. Although they are not in the top cidentally began to cut into a large number of patents in the
10, they are also listed in the table due to their influence (see field of speech recognition, speech, and digital information
Table 9). conversion in 2019. The clues can also be seen from their
IBM’s patents began to grow rapidly in 2016, when IBM’s products. The Google Nest Mini launched in November 2019
patents were concentrated in the two categories of G06F 17/ and the Apple HomePod launched in August 2019 show the
30 and G06F 17/27, showing that IBM focused on infor- development path from smart speaker to smart home. With
mation retrieval and grammar analysis in NLP. In 2019, the the maturity of natural language technology and IoT, the use
number of patents of Microsoft, Amazon, Accenture, and of natural language to control objects around life will
Univ Kunming Science and Tech began to grow significantly. gradually replace the previous method of operating through
In addition to G06F 12/27, Amazon and Microsoft use buttons or operating with limited system interfaces. When
speech recognition technology based on natural language other companies focus on deepening NLP-related tech-
models in human-machine dialogue, which is mainly re- nologies or developing speech recognition applications,
flected in the two IPCs G10L 15/18 and G10L 15/22. In 2020, Facebook Inc. has paid more attention to electric commu-
the number of patents of Google, Samsung, and Baidu nication technique, including H04L 12/58 and H04 29/08.
16 Complexity

Table 9: Top 10 assignees.

Top Assignee 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Total
1 IBM 12 14 22 24 26 89 129 231 313 498 1,358
2 Microsoft Technology Licensing LLC 0 0 0 0 12 14 23 39 72 125 285
3 Amazon Tech Inc. 0 2 0 0 8 16 15 35 64 57 197
4 Google LLC 0 1 0 0 0 0 0 15 36 133 185
5 Samsung Electronics Co. Ltd. 2 4 4 3 5 6 8 19 27 74 152
6 Nuance Communications Inc. 4 4 9 14 12 12 17 15 15 23 125
7 Accenture Global Solutions Ltd. 0 0 0 0 0 0 2 7 37 77 123
8 Beijing Baidu Netcom SCI & TEC 0 0 0 0 0 0 1 6 10 92 109
9 Microsoft Corp 21 21 14 17 8 2 4 3 2 2 94
10 Univ Kunming Science and Tech 0 1 0 0 1 3 1 15 47 23 91
16 Apple Inc. 0 1 0 3 1 1 6 10 19 36 77
17 Facebook Inc. 0 0 0 2 1 3 5 4 19 38 72

The two IPC codes represent message switching systems and compared with the patent text is extracted from Wikipedia.
transmission control procedure in network communication, Speech recognition, NER, NLU, and NLG are technologies
respectively. in the domain of NLP. Feature engineering, RNN, CNN, and
transformer are of model. And speech-generating device,
5. Technology Function Matrix cloud computing, voice activity detection, human-computer
interaction (HCI), and immersive technologies are of
A Technology Function Matrix (TFM), which investigates system.
the corresponding relation between technologies and
functions on patent amount, is a critical approach for patent
data analytics. The domain of NLP, model, and system, 5.2. Definition of Function. Nine TFM functions, which are
which is introduced before in Section 3.2.3, are used to form information extraction, dialogue management, context
the TFM. The construction process of TFM is described as prediction, recommendation system, algorithm efficiency,
the following. A well-constructed ontology is defined before, automated control, communication, user experience, and
from which technology and function terms can be defined, virtual assist, are listed in Table 11. The description of the
and patents can be collected by the search query set similarity compared with the patent text is extracted from
according to the ontology. Next, each patent is visited it- Wikipedia and other web resources.
eratively to count if it matches each technology and function.
By doing this, a TFM can be constructed.
5.3. TFM Result. For finding emerging trend of natural
This research uses the TF-IDF-based TFM automatic
construction method. After defining the technologies and language-enabled chatbot, year 2020 patents are used as the
functions, an unstructured text description that best rep- source for TFM. The 13 × 9 TFM result is obtained through
resents each technology or function must be prepared. These the automated process described before (see Table 12).
text descriptions are transformed into a set of vectors Transformer is a DL language model, developed in 2017,
widely used to process natural language tasks. The patents
through unsupervised learning, which acts as an agent for
related to transformer technology and prediction function
each technology or function. Then, specific fields are selected
are the highest number, which means transformer is a
from each patent, converted into a vector, and compared
mature technology and be widely applied for context
with each technology and function through similarity, and a
prediction. In terms of technologies (row), transformer and
threshold is used to determine whether the patent can be
speech-generating device are the main technologies of the
classified as the technology or function. Thus, the text de-
current market and have a positive impact on almost all
scription of each technology or function is very important.
Sections 5.1 and 5.2, respectively, explain the technologies functions. In terms of functions (column), automated
and functions selected in this study, followed by the TFM control function is more widely used than others. For
result in Section 5.3. After that, the domain of applied instance, speech recognition and speech-generating device
scenarios is added to form the three-dimensional matrix, are for increasing the pipeline of the control system. In
addition, the NLP domain technologies mostly relate to
which is called A-TFM and is introduced in Section 5.4.
information extraction, dialogue management, and pre-
diction, such as the improvement of NLU and NLG can
5.1. Definition of Technology. 13 TFM technologies, listed enhance the system’s ability to identify users’ intent. Last,
below in Table 10, are defined according to domain of NLP, the system domain technologies mostly concentrate on
model, and system. The description of the similarity communication, user experience, and virtual assistant. For
Complexity 17

Table 10: TFM technologies. In addition, speech recognition and automated control
ID Domain Technology
functions (F6) are combined with each other to form the
application of speech-driven automated control. When re-
T01 NLP Speech recognition
ceiving speech data from the client, speech recognition and
T02 NLP Named entity recognition
T03 NLP Natural language understanding NLU model stored in the cloud are used to interact with
T04 NLP Natural language generation other devices in the cloud space, such as unmanned aerial
T05 Model Feature engineering vehicles (UAVs), robots, augmented reality (AR), and virtual
T06 Model Recurrent neural network reality (VR) devices, through AI modules and 5G network
T07 Model Convolutional neural network technology.
T08 Model Transformer model
T09 System Speech-generating device
T10 System Cloud computing 5.3.2. NER (T02). In order to improve the accuracy of NER,
T11 System Voice activity detection preprocessing is very important. Patent CN110990525A
T12 System Human-computer interaction proposes a sentiment-based information extraction method
T13 System Immersive technologies that achieves good performance in the field of financial
sentiment information extraction through preprocessing
and feature extraction modules. Data labeling and feature
Table 11: TFM functions. engineering are the two main steps in preprocessing. Patent
ID Function CN111783466A proposes a named entity recognition
method for Chinese medical record field, in which the label
F1 Information extraction
F2 Dialogue management
uses two-layer conditional random field (CRF) classification
F3 Context to determine the final output label thus improving the ac-
F4 Recommendation system curacy of NER and reducing the time consumed by training.
F5 Algorithm efficiency There is similar research in literature studies. In view of the
F6 Automated control insufficient representation of potential features of Chinese
F7 Communication characters, Han et al. [44] uses the BiLSTM network to learn
F8 User experience the internal strokes and radical semantic information of
F9 Virtual assistant Chinese characters and combines with the BiLSTM-CRF
model to construct an adaptive multifeature fusion em-
bedded CNER model. In addition, patent WO2020167558A1
instance, adopting immersive technologies can enhance proposes a dynamically trained model of named entity
user experience or the development of cloud computing recognition over unstructured data, which defines entity
makes portable devices handling complex tasks. Therefore, labels for specific domain knowledge ontology, and uses
a lot of virtual assistants are developed to assist people for a these entity labels to identify the relationship between un-
convenient life, such as intelligent drive assistant. Next, the structured documents and domain knowledge. Patent
interaction of technology and function and its related CN111737969B proposes a resume analysis method based
patents are explored to find emerging technologies or on a DL model, which combines NLP, OCR, and named
applications. entity recognition technology. This method first performs
feature modeling on the resume. After the model training is
completed, the key information is classified and the category
5.3.1. Speech Recognition (T01). The most applied function
mapping model is set, so that the parser can read it like a
of speech recognition is information extraction (F3). Ac-
human and improve the overall analysis effect.
curacy of speech recognition is the key to determining
whether it can be applied to the commercial field, and good
information extraction ability is a necessary condition. 5.3.3. Transformer Model (T08). The transformer model is
Although speech recognition technology has gradually widely used to improve the accuracy of the information
matured, there are still a large number of patents in this field extraction function (F1). Patent CN110941698A proposes a
for better recognition capabilities and information extrac- method based on the bidirectional encoder representation
tion capabilities. on BERT CNN, which generates rich contextual semantic
Google LLC’s patent US10431206B2 uses the hierar- information of word vectors, thereby effectively supporting
chical recurrent neural network (HRNN) structure handles service similarity calculation to find the most accurate target
the task of multiaccent speech recognition. Patent service, and achieving accurate retrieval of target services.
CN110033766A proposes a complex multiple deep neural As for dialogue management function (F2), patent
network architecture, including single layer of one-way CN111274362A proposes a dialogue generation method
RNN model, binary bidirectional RNN model, and binary based on the transformer architecture, which involves
bidirectional LSTM (BiLSTM) model and other network obtaining a vectorized representation of words, and gen-
structure, in pursuit of faster speed and less energy con- erating a reply based on a comprehensive semantic vector
sumption. Patent EP3497630B1 uses CNN architecture, and a copy mechanism, which is used to solve the NLG based
which allows better signal propagation and long-range de- on background domain knowledge dialogue. Patent
pendency learning, thus improving output quality. US20200372341A1 proposes a pipelined natural language
18 Complexity

Table 12: The TFM result.

F1 F2 F3 F4 F5 F6 F7 F8 F9
T01 NLP Speech recognition 703 673 883 343 307 661 452 484 630
T02 NLP Named entity recognition 948 412 484 301 339 852 134 318 217
T03 NLP Natural language understanding 809 503 514 195 177 627 87 249 188
T04 NLP Natural language generation 989 724 827 308 348 759 105 378 272
T05 Model Feature engineering 571 377 306 436 372 470 136 299 190
T06 Model Recurrent neural network 569 386 613 257 244 332 190 110 161
T07 Model Convolutional neural network 317 251 446 287 327 208 149 98 144
T08 Model Transformer model 1,048 995 1,348 514 422 714 308 452 448
T09 System Speech-generating device 1,189 1,141 1,123 622 384 1,006 792 998 963
T10 System Cloud computing 341 422 237 358 213 465 487 452 458
T11 System Voice activity detection 309 377 241 283 120 379 780 557 705
T12 System Human-computer interaction 685 858 509 512 260 626 803 909 850
T13 System Immersive technologies 307 417 211 272 123 211 439 533 420

question answering system based on the BERT model, which maturity of Internet technology and mobile devices, the past
involves receiving an input text of a natural language information retrieval systems have begun to be replaced by
question and provides an answer to the natural language chatbots. However, when NLP technology is not yet mature,
question considering context. rule-based chatbots cannot exert influence. However, as NLP
The transformer model is used in context (F3) func- technology and speech recognition technology mature,
tion to improve the accuracy of NLP. Patent speech-generating devices have also developed rapidly and
CN110737764A proposes a method for generating per- combined with chatbot applications. Task-oriented re-
sonalized dialogue content based on a multiround dia- trieving systems began to be replaced by speech query
logue model. The transformer model effectively learns the systems. Patent CN110111766A claims a multifield multitask
dialogue sequence relationship between natural lan- system, which solves the problem of the multidomain
guages, can predict the generated content to reduce the multitask switching in the dialogue system. The complex
probability of replying commonality, and increase the multitask dialogue system integrates a speech recognition
diversity of dialogue content. Patent CN111708882A module, a domain confidence state tracking module, dia-
proposes a method for complementing missing Chinese logue managing module, an NLG module, and a speech
text information based on transformer encoder. This synthesis module to realize the capability that semantic level
method starts from manually preprocessing Chinese text information can be shared between each domain. Patent
documents, dividing the text into a large number of short JP2020098308A proposes a voice inquiry system for infor-
sentence corpora, and converting it into the smallest unit mation provision, in which each of chatbot servers and
of BERT vector. Since the purpose is to find out the smart speaker operation server use the DL model, accept a
missing words and sentences in the article, the training spoken question, infer, and output the corresponding an-
method is to randomly generate noise to hide the words in swer in spoken speech.
the complete article to create the effect of the omission. The next step after reaching the speech query system is
Conversely, in order to be able to fill in the missing words, speech-driven remote control. 1,006 patents related to au-
the model must have text generation capabilities. Through tomated control function also support this idea. Patent
repeated information deletion and generation procedures, US10748529B1 (assignee: Apple Inc.) proposes a voice-
Chinese natural language processing task accuracy is based digital assistant for use with home automation of voice
further improved. activated controllable device, such as TV, speaker, or
camera. The application of speech-driven automated control
is not so uncommon, but they are focused on devices that do
5.3.4. Speech-Generating Device (T09). Speech-generating not have safety hazards, such as home-related devices. It also
device is highly related to the three functions of information means that speech-driven automated control is still at the
extraction (F1), dialogue management (F2), and context auxiliary stage and cannot replace existing functions.
(F3), with 1,190, 1,141, and 1,123 patents, respectively. The However, it is believed that one day people will hope that
speech recognition technologies of T09 and T01 are also many functions that require physical contact can be replaced
highly related, but the classification of T09 in the “system” by voice control, and the first thing to overcome is noise.
domain means that the description of this technology is Since the sound is not specified, the device may receive
more focused on the hardware or system framework, so that unexpected sounds and trigger actions at any time. There-
for T09, F1, F2, and F3. The gap between is blurred. From fore, a gateway may be required to avoid unexpected actions
these large numbers of patents, it can be found that with the caused by noise. Patent US20140214414A1 proposes a
Complexity 19

communication system for use in automatic speech recog-

nition applications, which can transmit commands through
7
wireless network to modify gateway’s noise reduction

Applied scenarios
processing state. 6
5
4
5.3.5. HCI (T12). When it comes to smart homes, in ad- 3
dition to speech control, there are more automatic control 2
methods through HCI. Patent CN110932953A proposes s 1
smart home control method and device, which can receive 8 9
7
the user control command of the target home, login target 6
12 5 ons
start home residence in the target network, intelligently 10 3
4 c ti
8 n
perform control, and return the result message back. This 6 4 2 Fu
Techno 2 1
solution realizes the multihome for different manufacturers logies
and different communication protocols for uniform control. Personal Medical
It is observed from TFM that HCI technology is widely Ecommerce Engineering
used to improve user experience (F8), and there are 909 Education Driver assistant
patents located in the interaction. Most people use chatbots
to meet their needs, such as information retrieval or specific Society
operational tasks. It is most important to be able to meet the Figure 6: The three-dimensional matrix.
needs of users in fewer conversations. Many patents also aim
to reduce dialogue and improve dialogue efficiency, such as
CN112015879A, CN110990594A, CN111488433A, and 6.1. Knowledge Graph. AI makes huge progress; algorithms
CN110827831A. are rapidly improving, managing massive amount of data;
however, it still is not knowledge-driven technology. The
knowledge behind the natural language-enabled chatbot is
5.3.6. Immersive Technologies (T13) with Virtual Assistant very important for dialogue with humans. The early de-
(F9). In addition to the HCI methods of contact and voice, velopment of chatbot was mostly dominated by a single
the use of gaze tracking to help virtual assistants more ac- domain. It has been observed that more research has been
curately grasp the text or dialogue paragraph the user is directed towards open domain [45–49] and multidomain
paying attention to is an emerging application. [50–52] in recent years. Single-domain chatbots are limited
to accomplishing specific tasks, while multidomain or open
domain chatbots can better meet the needs of smart assis-
5.4. A-TFM and TFM with Applied Scenarios. As mentioned tants and even further provide people’s companionship or
in Section 3.4, the applied scenario factor is also a valuable social applications. With the development of 5G and cloud
part for analyzing patents. Therefore, this research utilizes applications combined with social media, many social
the applied scenarios as the third dimension to construct a 3- media, such as Telegram, Cortana, Slack, WeChat, Facebook
dimensional matrix. As shown in Figure 6, the scale of node Messenger, Google Assistant, and Siri, provide platforms
means the number of patents. X-axis means 13 technologies, that can easily build chatbots [53], making the transition of
Y-axis means 9 functions, and Z-axis means 7 applied technology bottleneck shifting from simple single-domain
scenarios. The source of this three-dimensional matrix is 50 chatbot system construction into complex integration of
patents which randomly collected from the source of the multidomain knowledge bases. The correlation between
above TFM. “Personal” and “e-commerce” are the main these two phenomena is hypothesized.
applied scenarios of the current market. “Medical,” “engi- With the rapid development of the semantic web, a large
neering,” and “driver assistant” are applied scenarios still amount of structured data has been provided in the form of a
under development. Also, few patents related to “education” knowledge based on the web. Making these data accessible
and “society” chatbots are found. and useful to end users is one of the main goals of chatbots
based on link data [54]. KG is considered to be a new AI
6. Discussion technology trend, which originated from the basic principles
of the Semantic Web and the construction of the knowledge
Nine topics, including medical data, smart cities, IoT, data base [55]. The novel KG-based framework is used in many
privacy, sustainable strategies, CRM, personalization, so- chatbot applications. They combine the query language
cial media listening, and ML models, are identified as latent SPARQL of the resource description framework to quickly
topics for future research based on data-driven strategies integrate the existing knowledge base.
[14]. This research thoroughly investigates the application Related patents in recent years have also focused on
of chatbots by comprehensive patent-mining process and studying how the knowledge framework can improve the
claims the consistency between the findings of this study capabilities of NLU and integrating the KG into the
and the above results. Thus, the effectiveness of proposed knowledge base of chatbot. Patent US10733375B2 (assignee:
analysis is justified. Apple Inc.) provides a system and process for operating
20 Complexity

intelligent automated assistants. This process is based on a standard queries using neural networks, with a processor
knowledge framework and can improve the validity of NLU, that determines the relevance of documents and returns
analyze the mapping of domain attributes and words from documents when they are determined to be relevant. This
the natural language input, then correspond to the data of application describes a system and method for converting
the knowledge base according to the analysis results, and natural language queries into standard queries using se-
determine the output response results according to the quence-to-sequence neural networks. As described in this
ranking mechanism. Patent EP3362972A1 proposed a sys- article, when a natural language query is received, the
tem for authoring visual representation for text-based natural language query is converted into a standard query
natural language document. User interface is provided that using a sequence-to-sequence model. In some cases, the
contains a document area and thus enables to interactively sequence-to-sequence model is associated with the layer of
generate the visual representation information that accu- interest. The perform searches using standard queries and
rately depicts the underlying source text. The system gen- can return various documents. The documents obtained by
erates a node graph of at least one of the parse trees, the the search are scored based at least in part on the deter-
entity information, or the relational phrase information and mined conditional entropy of the documents. Use natural
processes the document to determine relational phrase in- language queries and documents to determine conditional
formation indicating that the portion of the text includes a entropy.
relationship to at least one of a subject, verb, or object in a
sentence that includes the portion of the text. Also, the
system generates another visual representation links the 6.2. Deep Learning. The importance of algorithms related to
nodes and the relations. Patent WO2020160264A1 proposed AI and deep learning to chatbot is obvious. However, this
a method of identifying relevant data sets using training kind of emerging technology is less noticeable in patent
models related to topics of interest, involving access to one documents. Commonly used chatbots are LSTM, trans-
or more sources, each of which contains information sys- former, RNN, etc. Interestingly, the bidirectional mecha-
tems and related methods used to organize, represent, find, nism is applied to almost all architectures. Chatbot-related
discover, and access data. The embodiment represents in- articles using bidirectional architecture have appeared in
formation and data in the form of a data structure called a large numbers since 2019, and their number accounted for
“feature graph.” The feature graph includes nodes and edges, more than 80% of all years (see Table 13).
where edges are used to “connect” nodes to one or more Patent CN111267097A proposed a natural language-
other nodes. The nodes in the feature graph can represent based industrial robot-assisted programming method, in-
variables, that is, measured objects, features, or factors. The cluding parsing language instructions, matching analysis
edge in the feature graph may represent a measure of the results, and combining coordinate output to generate the
statistical association between a node and one or more other final robot-assisted code. The present invention requires a
nodes that have been retrieved from one or more sources. method for auxiliary programming of natural language-
The data set that represents or supports statistical correlation based industrial robots according to language instructions
or measurement correlation variables is “linked to” form the and generating corresponding executable codes for the
“feature graph.” Patent US10762113B2 (assignee: Cisco) environment image robot. The present invention is divided
proposes the use of conversational knowledge graphs in into three parts. First, use LSTM bidirectional recurrent
virtual assistants to process natural language input. After neural network (Bi-RNN) and fast regional convolutional
receiving the natural language query of the user, the method neural network (F-RCNN) to extract language instructions
retrieves the contextual information of the conversational and features of the factory environment. Second, provide the
knowledge according to the intention and calls the back-end “attention mechanism” model of the alignment algorithm,
service accordingly and obtains the response after the service and correctly match the machine translation of the in-
is performed. Finally, the response is translated into natural struction in the machine environment, so as to identify the
language and provided to the user. There are similar studies specified object and the output coordinate point of the
in literature studies. Zhong et al. [56] designed a cognitive object. Third, use the model output of the generating op-
information representation model based on the knowledge eration to match the CoBlox result modular programming
graph, which combines the perception information and model.
semantic description information of the industrial robot The technical development of DL in NLP has been quite
ontology to form a structured and logically reasoning mature. Although academic research is constantly pursuing
cognitive knowledge graph, including the perception layer better performance, it is already more than enough at the
and the cognitive layer. The realization of automatic rep- applied level. When applying any framework commonly
resentation of robot perception information enhances the used today, even with little training data, a chatbot is able to
versatility, systematicness, and intuitiveness of robot cog- be perceived satisfactory by users [57]. Therefore, in addition
nitive information representation and can effectively im- to being used to handle NLP tasks, the other main appli-
prove the cognitive reasoning ability and knowledge cation of DL is to assist the dialogue management of the
retrieval efficiency of robots in the industrial Internet chatbot system.
environment. Patent CN108282587B proposes a mobile customer
Patent US20200317093A1 proposed a query response service dialogue management method based on state
system for converting natural language queries into tracking and policy orientation for communication industry,
Complexity 21

Table 13: Bidirectional related article number. 6.4. Speech-Driven Automated Control. Interactive Smart
No. of results
Agents (ISAs), which are controlled by users through natural
Search terms language dialogues, are becoming a part of life, especially in
All Since 2019
smart home scenarios [62]. Patent WO2020203067A1 de-
Chatbot Bi-RNN 30 25 scribes an information-processing device containing a
Chatbot BiLSTM 422 347
control unit driven by natural language, which is arranged
Chatbot BERT 1,290 1,150
for controlling the movement of a moving object on the basis
of results of a speech recognition process. Patent
involves adopting the deep Q-network-based strategy op- CN110654738A describes an automatic garbage classifica-
timization method to select best action strategy. The method tion and recycling device based on NLP. The garbage bins
involves establishing a dialogue problem guiding strategy are, respectively, equipped with infrared sensors, and the
based on the partially observable Markov decision process lower box body is equipped with a mechanical transmission
(POMDP) model, and applying an action to dialogue en- mechanism and an automatic classification mechanism. The
vironment state of user through the internal action of the device and method of the present invention have high
POMDP model, so that the state of the conversation en- recognition efficiency and high degree of automation.
vironment changes and a certain return is obtained. The
likelihood of executing a series of strategies is measured
based on the cumulative returns obtained, and the problem 6.5. Internet of Things (IoT). Patent KR2020131299A (as-
is turned into a strategy choice problem. A deep-enhanced signee: Google LLC) proposes a method of associating
learning problem-guided strategy optimization algorithm is multiple remote automation assistant components through
constructed based on the dialogue problem guiding strategy IoT devices, combined with voice recognition modules to
obtained by the POMDP model, and a deep Q-network monitor and send voice data. Patent US10543931B2 pro-
(DQN)-based strategy optimization method is adopted to poses a method for monitoring audible and message alerts
select the best action strategy. received during flight in the aircrafts. IoT cockpit includes
subsequently marking a cascaded message alert to associate
with the display element. After receiving a plurality of alerts,
6.3. Speech-Related Technologies. Chatbot has developed including at least one of the audible alerts or message alarm,
towards an integrated conversation system, where in the the first NLP task is applied to convert the auditory alarm
context of multiperson conversations, speech segmentation into a text alarm that is structurally consistent with the
and speaker recognition algorithms have been the main format for aggregation, or a cascaded message alarm, where
research topics in recent years [58, 59]. Li et al. [60] sum- the second NLP task is applied to identify the context.
marizes the modern noise-robust technology of ASR de-
veloped in the past 30 years and proposes the classification
standards for various noise-robust technologies, and the 6.6. Applied Scenarios. According to the A-TFM results in
pros and cons of using different antinoise ASR technologies Section 5.4, it can be found that the related patents of chatbot
in actual application scenarios. For example, for stable voice- applications are still mainly focused on personalized services
controlled driving, the environmental conditions of drones and e-commerce. Both types of applications are focused on
must be handled carefully, including environmental noise using chatbot as a virtual assistant serving a specific purpose, or
that can reduce the accuracy of recognition. So, Park and Na using chatbot as an expert in a specific field to achieve the
[61] studied multiple unmanned aerial vehicle (UAV) purpose of knowledge acquisition. These applications for
control and noise reduction methods driven by voice. providing utility or productivity are progressing towards ed-
Patent CN111768768A proposes a method of processing ucation [63, 64], medical [65], emotional [66, 67], and social
voice in the fields of AI, DL, NLP and voice interaction, and services [68–70]. Under these conditions, the integration of
noise reduction processing on voice data sent by peripheral socioemotional behavior and personality processing design
control equipment. The specific implementation scheme is principles can lead to a decisive competitive advantage [71].
as follows: in response to the acquired voice recognition The application trend of chatbot obtained from the patent
interface call request sent by the peripheral control device, analysis in this study is consistent with some studies [71, 72],
start the voice recognition process; acquire the type of the which illustrates the effectiveness of this research.
peripheral control device; determine the target voice noise
reduction mode according to the type of the peripheral 7. Conclusion
control device. In the noise mode, noise reduction is per-
formed on the voice data sent by the peripheral control The study conducts a comprehensive patent review on
device to obtain the voice data after noise reduction; after emerging technologies of natural language-enabled chat-
noise reduction, voice recognition is performed on the voice bots. The contribution of this study is addressed in Section
data to generate text data. Therefore, through the voice 7.1, the managerial implication is described in Section 7.2,
processing method, the noise level generated by other op- the practical/social implications for marketers are described
erations in the peripheral control device included in the in Section 7.3, and the limitations and future research are
voice data is reduced. suggested in Section 7.4.
22 Complexity

7.1. Contribution. The contribution of this study is from the chatbots of various specific domains integrate with each
three aspects. First, a patent analytic framework is proposed other to become a more complete and powerful system.
and proved to be effective. Second, emerging technologies Communication technology and system integration are
are found. Third, application trend is addressed. also very important.
A patent analytic framework starts from patent-based As for the application trend, the increasing number of
ontology construction, followed by patent management map patents shows the rapid development of NLP chatbot in
and TFM, and performing the case study part. The four-level recent years. From the macroscopic patent trend analysis,
hierarchical structure of the ontology is constructed with the development trend of patents has been found. The
text-mining approaches such as k-means clustering algo- patents related to natural language-enabled started in 2014
rithm and LDA topic modeling, to reduce human inter- and developed rapidly since 2016. At first, it was mainly
ference during the process. The ontology map can be used as based on NLP and knowledge base. By 2018, speech rec-
the basis for strategic and sustainable R&D planning, from ognition and communication technology have been de-
which researchers are able to quickly understand the de- veloped and perfected, and then a large number of
velopment trends of key technologies and can identify applications began to appear in 2019. These applications are
technology gaps. It is worth noting that in some past patent concentrated in Silicon Valley’s technology giants, and they
analysis articles, detailed patent query conditions were first have also brought significant improvements to people’s
designed, on which the following analysis are based [25]. lives. Natural language-enabled chatbot is widely used in
However, the patent analysis method proposed in this re- the field of e-commerce, focusing on customer service and
search uses iterative process to find out the most appropriate medical consulting. With the popularization of 5G network
query conditions and patent information during the con- technology, more and more voice-driven applications, such
struction of ontology. In addition to patent analysis, it is as speech-driven automated control for IoT and system
reasonable to find emerging technologies from academic integration, along with immersive human-computer in-
articles, and systematic literature review (SLR) is the main teraction interfaces provide better user experience. In
method. Preferred Reporting Items for Systematic Reviews addition to e-commerce applications, more applications in
and Meta-Analysis (PRISMA) was created by the interna- the product life cycle process have begun to be observed.
tional health collaborator network and provides a frame- The application scenarios of natural language-enabled
work for the SLR to ensure methodological rigor and quality chatbot have clearly begun to shift from e-commerce to
[73]. The conduct of an SLR largely depends on the scope engineering applications, such as product design, engi-
and quality of the included research. Therefore, the sys- neering assets management, smart manufacturing, and
tematic reviewer may need to modify its original evaluation workshop management. Natural language-enabled chatbot,
plan in the process, and the PRISMA statement recognizes as an emerging smart system architecture using AI, has
this iterative process [74]. This provides crucial support for become a service integration solution through the inte-
the iterative method used in this study to continuously gration of devices, algorithms, and network communica-
adjust the patent query conditions in the ontology con- tion technologies. It is also expected to continue to impact
struction process. the traditional information system architecture in the
The emerging technologies are summarized as follows. future.
Knowledge is the basis of natural language-enabled chat-
bot, among which feature graph is a feature generation
framework that has recently attracted attention. DL is the 7.2. Managerial Implication. At present, the application of
core of the main method, and most of the DL algorithms chatbot is still focused on personal assistants and customer
are mature. In recent years, patents have focused on the services, and these application scenarios are limited to a very
combination of various DL algorithms, by capturing their limited field of knowledge
respective advantages and filling each other’s shortcom- From the early rule-based dialogue interaction system
ings. In terms of speech technology, noise reduction is the to natural language interaction, coupled with the maturity
focus of recent speech recognition technology. Sounds of voice recognition technology, chatbot can provide good
including voices and noise in operating equipment are dialogue quality in chit-chat and single-round dialogue.
obtained from the device and converted into refined text The bottleneck of service provision has shifted from system
data through the integration of DL and NLP technologies. development to the establishment of in-depth domain
Furthermore, it is found that context is the main research knowledge base. Many Internet service providers have
subject, whether it is the exploration of the knowledge base been able to provide a convenient application framework
or the logic of the algorithm. Previous research on NLP has for establishing chatbot as an automated customer service
focused on unstructured text, but in recent years, it has or personal service assistant. The success of the chatbot
clearly turned to messages in dialogue. In unstructured service depends on whether it accurately interprets users’
texts, the term frequency-based method can have good context or intented question and possesses the knowledge
results, but the message in the dialogue relies on a large base needed to fully support the context and provide
number of pronouns and the continuity and relevance of accurate replies.
the context, and the anaphora is more complicated. Even to The limitation of chatbot’s focus on a single domain has
be able to apply NLP to daily conversations, it faces a larger begun to be noticed, so the practice of integrating multiple
and broader domain and knowledge base. For this reason, domain chatbot into a chatbot advisory group has been seen
Complexity 23

in recent patents and research. With the changes in chatbot should be more convenient for human-computer inter-
system structure, multiple domain knowledges are inte- actions. Although studies have shown that consumers
grated into a complex system. In recent years, the strategy generally prefer to interact with people compared to
of focusing on data-driven innovation has led to new chatbots, giving human qualities can still effectively en-
products and business models in the emerging and devel- hance the consumer experience [81]. For marketers, it will
oping digital markets. However, while exploring knowledge be an important issue to strike a balance between com-
from data, user privacy is an issue that needs to be treated petent tasks and anthropomorphic enthusiastic
with caution [75, 76]. responses.
To sum up, the feature of chatbot shifts from simple
information provision to complex information integration
and versatile decision supports, which means the reasoning 7.4. Limitations and Future Research. The first limitation is
and automatic dialogue and interface controls must be that the data source selected for this study is patent docu-
addressed. Patents on the control of electronic devices for ments from the DI collective global database
smart homes or cars also support this idea. The smart search feature of the DI database uses natural
language processing and deep learning methods to help find
related patents that match the user’s domain description.
7.3. Practical/Social Implications for Marketers. The three Compared with the traditional field search, this is a great
main motivations of chatbot usage imply the importance feature that can help identify related patents faster and more
of social media to the development of chatbot, the po- accurately. Nonetheless, this limits the use of paid DI database
tential of chatbot, and immersive technology in the en- for comprehensive patent set. The second limitation is that
tertainment industry, and the issues of chatbot even though data-driven ontology construction methods are
implementation [72] investigated in this study, domain experts are still needed to
As a platform for people to initiate conversations, social be involved in the entire operation of the framework for two
media has become main chatbot interface applications to the main purposes, key term extraction and result verification.
end users. The rapid integration of social media and chatbot When searching for patents in a specific domain, relevant
in e-commerce sites continues to grow and evolve. term will appear in a large number of patent documents.
The second most important application motivation is Although the TD-IDF vectorization mechanism has con-
entertainment, which is rarely addressed in patent docu- sidered both the number of terms and the uniqueness in all
ments. The realism of chatbot is still insufficient, but it can documents, the clustering results show that each cluster still
already provide rich and interesting interaction. In terms of contains a large number of common terms. In the results of
industrial development process, VR is at a similar stage. The topic modeling, these general terms are the main topics
VR experience itself is very attractive, just like an exciting corresponding to the clustering results, which indirectly
game, so the user experience when creating a virtual envi- confirms the validity of the method of this research. However,
ronment is far more important than the degree of realism even though we construct ontology from patent documents
[77]. It can also be found from the results of TFM that there through a data-driven method, we still need domain experts
are some patents located in chatbot combined with to verify the correctness of its ontology. In addition, in the
immersive technology to improve user experience. For construction process of TFM, this research also explores the
digital marketers, it implies that combining VR and chatbot scenarios in which these technologies and functions are ap-
in marketing and entertainment is expected to bring users a plied. Terms related to these scenarios are mentioned in
more immersive and innovative experience. patent data but occupy little number of words. This is also a
The third most application motivation is about social limitation on TF-based text-mining method.
services, such as social care for the elderly living alone. In Future research will solve the problems mentioned
the 3D-TFM proposed in this research, some patents for above. The first is to expand the source of data. In addition to
chatbot applications in social services and education patent data, Ribeiro-Navarrete et al. [76] proposed an SLR
scenarios have indeed been observed. The Turing Test was method of analyzing academic articles or the nonpatent
proposed in 1950 as a method to examine how a machine literature. It is expected that a more comprehensive view
behaves like a person [78]. In 2000, 50 years later, there might be provided by adding SLR in future research, and the
has been a lot of controversy about the relationship be- comparison between the results of SLR and patent-mining
tween the Turing Test and AI development [79]. However, can be further investigated. Moreover, how to better elim-
now, with the mature development of DL technology inate repeated terms in unstructured documents iteratively
nowadays that brings clear productivity and benefits, it is or other approaches will help to make text-mining methods
not that important whether a chatbot behaves like a more focused on finding unique representing terms in
person. An article on the application of chatbot in health specific domain. Thus, since quantitative and similarity-
care also mentioned that “AI needs to pass the imple- based text-mining approaches have been applied and reach
mentation game, not the imitation game” [80]. The ap- the limit, advanced technologies related to key term iden-
plications of service industries, such as entertainment, tification are clearly very important future research. Despite
social service, and education, imply that chatbot should the above limitations, the framework proposed in this study,
not be regarded as merely an emulated person, but a which analyzes the development of natural language-
system interface that can talk in natural language and enabled chatbot with quantitative supporting data, finds
24 Complexity

emerging technologies and points out possible future de- [10] M. Baez, F. Daniel, F. Casati, and B. Benatallah, “Chatbot
velopment directions and is still comprehensive and effec- integration in few patterns,” IEEE Internet Computing, vol. 99,
tive. In addition, this method and framework are universal 2020.
and can be easily applied to discover emerging technologies [11] L. Goasduff, Chatbots Will Appeal to Modern Workers, Gartner,
in other domains. Inc., Stanford, CT, USA, 2019, https://www.gartner.com/
The patent analysis method proposed in this research is smarterwithgartner/chatbots-will-appeal-to-modern-workers.
[12] C.-C. Chang and C.-Y. Liu, “Using patent deployment to
used to explore the emerging technologies and trends of
support industrialization of new technology,” Intellectual
natural language-enabled chatbot, which can reach high Property Rights Monthly, vol. 224, pp. 6–21, 2017.
consistency with the hints given in academic research. The [13] Y. Keng, Application of Patent Information to Business
methodology of this research is not restricted by a specific Planning of Enterprises, Judicial Yuan, Taipei, Taiwan, 2019,
domain, so the authors hope that this methodology can be https://www.judicial.gov.tw/tw/cp-1429-66877-ae6e7-1.html.
used as a reference for researchers to explore more emerging [14] J. R. Saura, “Using data sciences in digital marketing:
technologies and trends in other fields, so as to demonstrate framework, methods, and performance metrics,” Journal of
the contribution of this research. Innovation & Knowledge, vol. 6, 2020.
[15] B. Yoon and Y. Park, “A text-mining-based patent network:
Conflicts of Interest analytical tool for high-technology trend,” The Journal of High
Technology Management Research, vol. 15, no. 1, pp. 37–50,
The authors declare that they have no conflicts of interest. 2004.
[16] A. Abbas, L. Zhang, and S. Khan, “A literature review on the
state-of-the-art in patent analysis,” World Patent Information,
Acknowledgments vol. 37, 2014.
[17] Y. G. Kim, J. H. Suh, and S. C. Park, “Visualization of patent
This research was partially supported by research grant
analysis for emerging technology,” Expert Systems with Ap-
funded by the Ministry of Science and Technology (grant no. plications, vol. 34, no. 3, pp. 1804–1812, 2008.
MOST-108-2221-E-007-075-MY3). The authors also express [18] G. Kim and J. Bae, “A novel approach to forecast promising
their gratitude to Yi-An Su for helping refine the illustrations technology through patent analysis,” Technological Forecast-
in the paper. ing and Social Change, vol. 117, pp. 228–237, 2017.
[19] M. Thilakaratne, K. Falkner, and T. Atapattu, “A systematic
References review on literature-based discovery workflow,” PeerJ Com-
puter Science, vol. 5, p. e235, 2019.
[1] E. Ntoutsi, P. Fafalios, U. Gadiraju et al., “Bias in data-driven [20] U. H. Govindarajan, A. Trappey, and C. Trappey,
artificial intelligence systems—An introductory survey,” “Immersive technology for human-centric cyberphysical
WIREs Data Mining and Knowledge Discovery, vol. 10, no. 3, systems in complex manufacturing processes: A compre-
p. e1356, 2020. hensive overview of the global patent profile using collective
[2] L. Goasduff, 2 Megatrends Dominate the Gartner Hype Cycle for intelligence,” Complexity, vol. 2018, Article ID 4283634,
Artificial Intelligence, Gartner, Inc., Stanford, CT, USA, 2020, 14 pages, 2018.
https://www.gartner.com/smarterwithgartner/2-megatrends-do [21] V. Singh, K. Chakraborty, and L. Vincent, “Patent data-
minate-the-gartner-hype-cycle-for-artificial-intelligence-2020/. base: Their importance in prior art documentation and
[3] L. Wood, Global Chatbot Market Anticipated to Reach $9.4 Billion patent search,” Journal of Intellectual Property Rights,
by 2024 - Robust Opportunities to Arise in Retail & eCommerce, vol. 21, 2016.
Insider Inc., New York, NY, USA, 2019, https://markets. [22] O. Krejcar, R. Frischer, R. Hlavica, K. Kuca, P. Maresova, and
businessinsider.com/news/stocks/global-chatbot-market-anticipat A. Selamat, “Review of available SW solutions for intellectual
ed-to-reach-9-4-billion-by-2024-robust-opportunities-to-arise-in- property management systems from the perspective of open
retail-ecommerce-1028759508. innovation,” Journal of Open Innovation: Technology, Market,
[4] A. Chopra, 21 Vital Chatbot Statistics for 2020, OutGrow, New and Complexity, vol. 6, no. 2, p. 23, 2020.
York, NY, USA, 2020, https://outgrow.co/blog/vital-chatbot-
[23] H. Chiu, Smart Search, Clarivate, Philadelphia, PA, USA,
statistics.
2021, https://clarivate.com.tw/blog/2017/11/08/derwent-
[5] R. Dale, “The return of the chatbots,” Natural Language
innovation-ai-smart-search/.
Engineering, vol. 22, no. 5, pp. 811–817, 2016.
[24] S.-S. Weng, H.-J. Tsai, S.-C. Liu, and C.-H. Hsu, “Ontology
[6] A. S. Gillis, Turing Test, TechTarget, Newton, MA, USA, 2019,
https://searchenterpriseai.techtarget.com/definition/Turing- construction for information classification,” Expert Systems
test. with Applications, vol. 31, no. 1, pp. 1–12, 2006.
[7] S. Reeves, V. Williams, F. M. Costela et al., “Narrative video [25] A. J. C. Trappey, C. V. Trappey, and A.-C. Chang, “Intelligent
scene description task discriminates between levels of cog- extraction of a knowledge ontology from global patents,”
nitive impairment in Alzheimer’s disease,” Neuropsychology, International Journal on Semantic Web and Information
vol. 34, no. 4, pp. 437–446, 2020. Systems, vol. 16, no. 4, pp. 61–80, 2020.
[8] J. Dai and Z. Ma, “Automatic identification of bond infor- [26] D. Tsatsou, P. C. Davis, J. Li, I. Kompatsiaris, and
mation based on OCR and NLP,” Journal of Computers, S. Papadopoulos, Ontology Construction, Google Technology
vol. 14, no. 6, pp. 397–403, 2019. Holdings LLC, Mountain View, CA, USA.
[9] V. K. Jain and S. Kumar, “Predictive analysis of emotions for [27] R. Subhashini and J. Akilandeswari, “A survey on ontology
improving customer services,” in Natural Language Pro- construction methodologies,” International Journal of En-
cessing: Concepts, Methodologies, Tools, and Applications, terprise Computing and Business Systems, vol. 1, no. 1,
pp. 808–817, IGI Global: Hershey, PA, USA, 2020. pp. 60–72, 2011.
Complexity 25

[28] Y.-H. Tseng, C.-J. Lin, and Y.-I. Lin, “Text mining techniques [45] D. Adiwardana, M.-T. Luong, D. R. So et al., “Towards a
for patent analysis,” Information Processing & Management, human-like open-domain chatbot,” 2020, http://arxiv.org/
vol. 43, no. 5, pp. 1216–1247, 2007. abs/2001.09977.
[29] G. Battsengel, S. Geetha, and J. Jeon, “Analysis of techno- [46] S. Roller, E. Dinan, N. Goyal et al., “Recipes for building an
logical trends and technological portfolio of unmanned aerial open-domain chatbot,” 2020, http://arxiv.org/abs/2004.13637.
vehicle,” Journal of Open Innovation: Technology, Market, and [47] S. Bao, H. He, F. Wang et al., “Plato-2: Towards building an
Complexity, vol. 6, no. 3, p. 48, 2020. open-domain chatbot via curriculum learning,” 2020, http://
[30] D. Korobkin, S. Fomenkov, A. Kravets, and S. Kolesnikov, arxiv.org/abs/2006.16779.
“Methods of statistical and semantic patent analysis,” in [48] S. S. Abdullahi, S. Yiming, A. Abdullahi, and U. Aliyu, “Open
Proceedings of the Conference on Creativity in Intelligent domain chatbot based on attentive end-to-end Seq2Seq
Technologies and Data Science, Springer, Volgograd, Russia, mechanism,” in Proceedings of the 2019 2nd International
September 2017. Conference on Algorithms, Computing and Artificial
[31] J. Hu, S. Li, Y. Yao, L. Yu, G. Yang, and J. Hu, “Patent keyword Intelligence, Sanya, China, December 2019.
extraction algorithm based on distributed representation for [49] C. H. Hong, Y. Liang, S. S. Roy et al., “Audrey: A personalized
patent classification,” Entropy, vol. 20, no. 2, p. 104, 2018. open-domain conversational bot,” 2020, http://arxiv.org/abs/
[32] T. Shanie, J. Suprijadi, and Zulhanif, “Text grouping in patent 2011.05910.
analysis using adaptive K-means clustering algorithm,” in [50] P. Rastogi, A. Gupta, T. Chen, and L. Mathias, “Scaling multi-
Proceedings of the AIP Conference 2017, AIP Publishing LLC, domain dialogue state tracking via query reformulation,”
Bikaner, India, November 2017. 2019, http://arxiv.org/abs/1903.05164.
[33] S. Li, J. Hu, Y. Cui, and J. Hu, “DeepPatent: patent classifi- [51] D. Calvaresi, J.-P. Calbimonte, E. Siboni et al., “EREBOTS:
cation with convolutional neural networks and word em- privacy-compliant agent-based platform for multi-scenario
bedding,” Scientometrics, vol. 117, no. 2, pp. 721–744, 2018. personalized health-assistant chatbots,” Electronics, vol. 10,
[34] J.-S. Lee and J. Hsiang, “Patent classification by fine-tuning no. 6, p. 666, 2021.
BERT language model,” World Patent Information, vol. 61, [52] C.-Y. Li, D. Ortega, D. Väth et al., “ADVISER: A toolkit for
Article ID 101965, 2020. developing multi-modal, multi-domain and socially-engaged
[35] S. Jun, “Technology integration and analysis using boosting conversational agents,” 2020, http://arxiv.org/abs/2005.01777.
and ensemble,” Journal of Open Innovation: Technology, [53] N. A. Ahmad, M. H. Che, A. Zainal, M. F. Abd Rauf, and
Market, and Complexity, vol. 7, no. 1, p. 27, 2021. Z. Adnan, “Review of chatbots design techniques,” Interna-
[36] Technology Function Matrix Analysis-Quickly Grasp the Patent tional Journal of Computer Applications, vol. 181, no. 8,
Layout and Explore the Blue Ocean Opportunities, Wideband IP pp. 7–10, 2018.
Office, Taipei, Taiwan, 2017, http://www.widebandip.com/tw/ [54] A. Ait-Mlouk and L. Jiang, “KBot: a Knowledge graph based
mobile/knowledge2.php?type1�B&idno�224. chatBot for natural language understanding over linked data,”
[37] A. J. C. Trappey, C. V. Trappey, U. H. Govindarajan, and
IEEE Access, vol. 8, pp. 149220–149230, 2020.
A. C. C. Jhuang, “Construction and validation of an ontology- [55] P. A. Bonatti, S. Decker, A. Polleres, and V. Presutti,
based technology function matrix: technology mining of cyber
Knowledge graphs: New directions for knowledge representa-
physical system patent portfolios,” World Patent Information,
tion on the semantic web (dagstuhl seminar 18371)Schloss
vol. 55, pp. 19–24, 2018.
Dagstuhl-Leibniz-Zentrum fuer Informatik, Wadern, Ger-
[38] Y. T. Demey and D. Golzio, “Search strategies at the European
many, 2019.
patent office,” World Patent Information, vol. 63, Article ID
[56] D. Zhong, Y.-A. Zhu, L. Wang, J. Duan, and J. He, “A
101989, 2020.
cognition knowledge representation model based on multi-
[39] Y. Yang and G. Ren, “HanLP-based technology function
matrix construction on Chinese process patents,” Interna- dimensional heterogeneous data,” Complexity, vol. 2020,
tional Journal of Mobile Computing and Multimedia Com- Article ID 8812459, 17 pages, 2020.
munications, vol. 11, no. 3, pp. 48–64, 2020. [57] N. Tavanapour and E. A. Bittner, “Automated Facilitation for
[40] W. Ki and K. Kim, “Generating information relation matrix Idea Platforms: Design and Evaluation of a Chatbot Proto-
using semantic patent mining for technology planning: a case type,” in Proceedings of the Thirty Ninth International Con-
of nano-sensor,” IEEE Access, vol. 5, pp. 26783–26797, 2017. ference on Information Systems, San Francisco, CA, USA,
[41] A. J. C. Trappey, C. V. Trappey, T. M. Wang, and 2018.
M. Y. L. Tang, “Ontology-based technology function matrix [58] Z. Su, Y. Li, and G. Yang, “Dietary composition perception
for patent analysis of additive manufacturing in the dental algorithm using social robot audition for Mandarin Chinese,”
industry,” International Journal of Manufacturing Research, IEEE Access, vol. 8, pp. 8768–8782, 2020.
vol. 12, no. 1, pp. 64–82, 2017. [59] Q. Li, R. Gravina, Y. Li, S. H. Alsamhi, F. Sun, and G. Fortino,
[42] Analysis of Common Reasons for Rejection of Taiwan’s Arti- “Multi-user activity recognition: challenges and opportuni-
ficial Intelligence-Related Patents, IP Office, Ministry of ties,” Information Fusion, vol. 63, pp. 121–135, 2020.
Economic Affairs, Taipei, Taiwan, 2019, https://www.tipo.gov. [60] J. Li, L. Deng, Y. Gong, and R. Haeb-Umbach, “An overview
tw/tw/cp-85-859330-1189b-1.html. of noise-robust automatic speech recognition,” IEEE/ACM
[43] Everything You Need to Get Started in Medical Billing & Transactions on Audio, Speech, and Language Processing,
Coding, Medical Billing and Coding.org, a Red Ventures vol. 22, no. 4, pp. 745–777, 2014.
Company, Indian Land, SC, USA, 2021, https://www. [61] J.-S. Park and H.-J. Na, “Front-end of vehicle-embedded
medicalbillingandcoding.org/medical-billing-coding/. speech recognition for voice-driven multi-UAVs control,”
[44] X. Han, F. Zhou, Z. Hao et al., “MAF.-CNER:A Chinese Applied Sciences, vol. 10, no. 19, p. 6876, 2020.
named entity recognition model based on multifeature [62] G. Spinelli, U. Shamim, A. Woodcock, and A. Nair, “Enabling
adaptive fusion,” Complexity, vol. 2021, Article ID 6696064, by Voice. Voice Enabled Environmental Control (EC) Devices
9 pages, 2021. Using Interactive Smart Agents,” in Proceedings of the Twenty-
26 Complexity

fifth Americas Conference on Information Systems, ISAs, [79] A. P. Saygin, I. Cicekli, and V. Akman, “Turing test: 50 years
Cancun, Mexico, 2019. later,” Minds and Machines, vol. 10, no. 4, pp. 463–518, 2000.
[63] G. Vladova, J. Haase, L. S. Rüdian, and N. Pinkwart, Edu- [80] J. Powell, “Trust me, I’ma chatbot: how artificial intelligence in
cational Chatbot with Learning Avatar for Personalization, health care fails the Turing test,” Journal of Medical Internet
2019. Research, vol. 21, no. 10, Article ID e16222, 2019.
[64] S. Gupta, K. Jagannath, N. Aggarwal et al., “Artificially In- [81] R. Roy and V. Naidoo, “Enhancing chatbot effectiveness: the
telligent (AI) Tutors in the Classroom: A Need Assessment role of anthropomorphic conversational styles and time
Study of Designing Chatbots to Support Student Learning,” in orientation,” Journal of Business Research, vol. 126, pp. 23–34,
Proceedings of the PACIS 2019, X’ian, China, 2019. 2021.
[65] F. Mehfooz, S. Jha, S. Singh, S. Saini, and N. Sharma, “Medical
chatbot for novel COVID-19,” in ICT Analysis and Appli-
cations, pp. 423–430, Springer, Berlin, Germany, 2021.
[66] F. Catania, N. Di Nardo, F. Garzotto, and D. Occhiuto,
“Emoty: an emotionally sensitive conversational agent for
people with neurodevelopmental disorders,” in Proceedings of
the 52nd Hawaii International Conference on System Sciences,
Maui, HI, USA, January 2019.
[67] M. Jaiswal, C.-P. Bara, Y. Luo et al., “Muse: a multimodal
dataset of stressed emotion,” in Proceedings of the 12th
Language Resources and Evaluation Conference, Marseille,
France, May 2020.
[68] A. Augello, M. Gentile, L. Weideveld, and F. Dignum, “A
model of a social chatbot,” in Smart Innovation, Systems and
Technologies, vol. 55, pp. 637–647, Springer, 2016.
[69] A. Xu, Z. Liu, Y. Guo, V. Sinha, and R. Akkiraju, “A new
chatbot for customer service on social media,” in Proceedings
of the 2017 CHI Conference on Human Factors in Computing
Systems, Denver, CO, USA, May 2017.
[70] T. Hu, A. Xu, Z. Liu et al., “Touch your heart: a tone-aware
chatbot for customer care on social media,” in Proceedings of
the 2018 CHI Conference on Human Factors in Computing
Systems, Montreal, QC, Canada, April 2018.
[71] A. Janssen, J. Passlick, D. Cordona, and M. Breitner, “Virtual
assistance in any context: a taxonomy of desgin elements for
domain-specific chatbots,” Business & Information Systems
Engineering, vol. 62, 2020.
[72] P. B. Brandtzaeg and A. Følstad, “Why people use chatbots,”
in Proceedings of the International Conference on Internet
Science, Springer, Florence, Italy, November 2017.
[73] D. Pati and L. N. Lorusso, “How to write a systematic review
of the literature,” HERD: Health Environments Research &
Design Journal, vol. 11, no. 1, pp. 15–30, 2018.
[74] D. Moher, A. Liberati, J. Tetzlaff, D. G. Altman, and
a. t. P. Group, “Reprint-preferred reporting items for sys-
tematic reviews and meta-analyses: the PRISMA statement,”
Physical Therapy, vol. 89, no. 9, pp. 873–880, 2009.
[75] J. R. Saura, D. Ribeiro-Soriano, and D. Palacios-Marqués,
“From user-generated data to data-driven innovation: a re-
search agenda to understand user privacy in digital markets,”
International Journal of Information Management, vol. 2,
Article ID 102331, 2021.
[76] S. Ribeiro-Navarrete, J. R. Saura, and D. Palacios-Marqués,
“Towards a new era of mass data collection: assessing pan-
demic surveillance technologies to preserve user privacy,”
Technological Forecasting and Social Change, vol. 167,
p. 120681, 2021.
[77] M. S. van Gisbergen, I. Sensagir, and J. Relouw, “How real do
you see yourself in VR? The effect of user-avatar resemblance
on virtual reality experiences and behaviour,” in Augmented
Reality and Virtual Reality, pp. 401–409, Springer, Berlin,
Germany, 2020.
[78] A. M. Turing, “I.-Computing machinery and intelligence,”
Mind, vol. LIX, no. 236, pp. 433–460, 1950.