International Conference on Communication and Signal Processing, July 28 - 30, 2020, India

Review on Implementation Techniques of

Chatbot
Nithuna S and Laseena C.A


Abstract—Chatbots are an intelligent system being developed created with NLP methods and machine learning algorithms.
using artificial intelligence (AI) and natural language processing Likewise, there are numerous business chat engines accessible,
(NLP) algorithms. It successfully interfaces and answers the in- which help assemble chatbots dependent on customer
quiries of the users. Dialogue/conversation operators are predom-
inately utilized by organizations, government associations, and
information input. As of late, there has been a significant
non-benefit associations. They are habitually conveyed by money increment in enthusiasm for the utilization and deployment of
related associations like a bank, charge card organizations, or- dialogue generation systems.
ganizations like online retail locations and new businesses. These Many significant tech organizations are utilizing a virtual
conversational specialists are utilized in organizations running assistant or chat agent to fill the necessities of clients. A
from exceptionally small start-ups to enormous partnerships. chatbot is a conversational specialist where a computer
There are numerous chatbot advancement systems accessible in
the market both code-based and interface based. However, they program intended to reproduce an intelligent discussion [1-7].
don’t have the handiness and adaptability in growing real Nowadays various types of communication agents are in-
dialogues. Among popular intelligent personal assistants includes vented. But Artificial Intelligence Markup Language (AIML)
“Amazon’s Alexa, Microsoft’s Cortana and Google’s Google based chatbot is the highest popular that they are simple to
Assistant”. The elements of these specialists are restricted. The arrange. One of the examples of AIML based chatbot is
current chatbots are created utilizing rule-based techniques,
ALICE.
simple machine learning algorithms or retrieval based techniques
but they do not generate good results. This paper provides a Recently in the conversational agent system, there has been
critical review of chatbots and the current strategies are much development and experimentation. The custom-ary
exhaustively explored and talked discussed. chatbots are created by utilizing rule-based strategies, or
straightforward machine learning calculations, many pro-
Index Terms—AIML, ALICE, Chatbot, DNN, Machine learn- gressed chatbots are utilizing propelled NLP methods and
ing and NLP Deep Learning Techniques like Deep Neural Network (DNN)
and Deep Reinforcement Learning (DRL). In encoder-decoder
architecture use Sequence to Sequence (Seq2Seq) model is
I. INTRODUCTION
very popular for dialogue generation, language modeling and

C ONVERSATIONAL operator or Chatbot is a program

that creates a reaction dependent on offered contribution
to imitate human discussions in content or voice mode. These
machine translation. Seq2Seq uses Recurrent Neural Net-
work(RNN) which is a popular Deep Neural Network ar-
chitecture, especially for Natural Language Processing tasks.
applications are intended to mimic human-human interactions. In Sequence to Sequence (Seq2Seq) model, many to many
Chatbots are predominantly used in business and corporate RNN architecture is used for the decoder. In this, encoder-
organizations. Their functioning can range from customer decoder architecture, the input sequence is fed as a vector
service, product suggestion, product inquiry to a personal representation of text to the encoder[8]. Then, some
assistant. Rule-based techniques, retrieval techniques or simple intermediate representation of information or thought vectors
machine learning algorithms are used to build these chat is produced by the encoder. accordingly, the thought vector
agents. In retrieval-based techniques, chat agents check for generated by the encoder is fed into a decoder as input.
keywords inside the information expression and recover Finally, the thought vector processes the decoder and converts
relevant answers depen-dent on the query string. They depend the sequence one by one word and produces multiple outputs
on keyword similarity and retrieved text from internal or from the decoder in the form of a target sequence. Though,
external data sources including World Wide Web or vanilla RNN is default in Seq2Seq and works well for many
organizational database. Some other progressed chatbots are NLP problems yet, due to higher complexity of language
modeling problem, vanilla RNN cells often fail, specially,
Nithuna S is associated with Department of Electronics and Communi- where long sequence of information needs to be remembered,
cation, Government College of Engineering Kannur, Kerala, India. (e-mail: as this information frequently becomes large for bigger
[email protected])
Laseena C.A is a faculty of Department of Electronics and Communi- datasets and turns to information bottleneck for the RNN
cation, Government College of Engineering Kannur, Kerala, India. (e-mail: network[9-12].
[email protected]).

978-1-7281-4988-2/20/$31.00 ©2020 IEEE 0157

Authorized licensed use limited to: University of Edinburgh. Downloaded on September 05,2020 at 02:21:18 UTC from IEEE Xplore. Restrictions apply.
 This paper is prepared as follows: in Section II, some papers The PBO mod-ule is based on CBR (Case-Based
on chatbot systems are reviewed and compared along with its Reasoning) and problem-oriented solving. In the CBR
challenges, followed by the concluding remarks in Section III. approach, a case-based reasoner recalls a similar past case and
its solution. Some reasoner at that point adjusts the fruitful
II. REVIEW OF LITERATURE arrangement from reviewed evidence to alter any contrasts
Chatbot is an AI software. It can communicate with a user in between the reviewed evidence including the present situation.
natural language through messaging applications, websites, At last, this CBR collects the answer for the present state
mobile apps. Conversation between human and an automated alongside criticism regarding this result with the goal that it
conversational agents are more challenging. In this paper, very well may be utilized in taking care of future issues. PBO
reviewed some literature that uses different techniques for module of Dorothy’s performances likewise.
chatbot implementation.
A. Using chatbots for network management training
through problem –based oriented ed-ucation
M. D. Leonhardt et al. presented an approach, a chatbot is a
program that simulates typed conversation, with the point of,
fooling a human into thinking the human is talking to someone
else. a chatbot is a communication software that simulates
typed or voice communication with the human, including the
purpose of fooling into thinking a human to talking to another
person. A network management-aware chat-bot is created
named as Dorothy, it manages the network by management
protocols so since SNMP.
The management structure such as SNMP contains four
elements: Network Management Station (NMS), managed
elements, Management Information Base (MIB), and the pro-
tocol is used for message exchange. Computers, routers, and Fig. 1. System Architecture.
terminal servers are managed objects. To have the option to
deal with those components, specialists are vital. Software The module is capable of recovering past obstacles and
modules are agents that remain in network elements. The solutions that are previously included in the database and
Management data, for example, The network element collects providing guidance to and encourage user thought and more
and stores the received error packets. The MIB holds any knowledge searching. Two distinct way works the PBO
database from oversaw things that remain at this operator. The module that is : (i) helping clients to discover an answer during
network management application is received by NMS. That issues all previously figured into this system(while utilizing
contains at least individual procedures that speak by some Dorothy to different inquiries or descriptions) and (ii)
operators within this system with giving commands and exhibiting issues furthermore, arrangements conceivable
receiving answers. outcomes by recovering them straightforwardly from the CBR
Fig. 1 shows an in-depth diagram of Dorothy’s design. it’s database.
supported ALICE resolution. Dorothy integrated by AL-ICE
on producing five parameters that denote topic, address, B. Humorist bot: bringing computational hu-mour in a
operation, time and situationm. users asked questions are chat-bot system
defined in the “topic”. The “address” represents the device IP A. Augello et al. [12] exhibited a methodology, Humorist
number when the user asks the question. The “operation” bot utilizes all around established computational humor sys-
defines in SNMP contains which set of actions to get the tems that utilizing ALICE structure inserted toward a Yahoo!
information. The “time” is used when a question involves the Messenger customer. Humorist bot also a chatbot furnished
behavior of the network in recent days and the “situation” among feeling of entertainment, this fit for advising clever
receives information about a specific word in the question, like over every client and this implies likewise equipped for
average, maximum or minimum. entertaining fun, also understand their entertaining level.
This framework comprises three modules: central module, Fig. 2 displays this structure design. The chatbot knowl-
PBO module, and network history information module. AL- edge base is given in to the framework center part, three
ICE created knowledge is received by the central module. varieties about AIML classes are:
collect module and history information module are included in 1. Standard Alice’s class arrangement permits the chatbot to
the network history information module. Information about the carry a comprehensive discussion by the client..
network and history information database is collected by the 2. Humorist sentence generation is permitted by a set of
collect module. Measurement of the network between intervals classes. This class allow a humoristic way to respond by
is collected by the history information base. the chatbot to the client.

0158

Authorized licensed use limited to: University of Edinburgh. Downloaded on September 05,2020 at 02:21:18 UTC from IEEE Xplore. Restrictions apply.
3. Some classes which permit the chatbot to perceive a
humoristic purpose toward the client sentences. Collect
this component associating with the information base of a
chatbot to outer assets, for example, the lexical lexicon
WordNet and CMU pronunciation dictionary, all these
recognize the behavior of humoristic phonetic includes
within this sentence.
1) Humoristic texts characteristics recognition: Ac-
knowledgment of humoristic writing highlights is in-
creasingly hard to deal with yet has more effect on the
last client. The principle humoristic highlights are
“alliteration”, “antinomy”, and “adult slang”.
i. “Alliteration”: The redundancy about a word,
num-ber or phonetic sound near progressive
information remains a component existing in a large
Fig. 2. Humorist Bot Architecture [12].
portion about the humoristic sentences. These
phonetic impact instigated by certain redundancies the assigned target. such a situation proposes unaided apriori
catches in each consideration of individuals hearing to like an algorithm that delivers subtasks and their substantial
it, the majority of situations, delivers every clever out- orders. So in each corpus to remove subtask, here design is
come. The sentence is at first prepared to evacuate dependable. In the first place, it does expect to group plus
accentuation marks and stopwords, at that point its canonicalize comparable expression semantically to supplant
phonetic translation is gotten utilizing the CMU whole name substances on their sort also in every expression
dictionary. to give feature vector. Bunched to group feature vectors se-
ii. “Antinomy”: Every nearness about antinomies mantically among comparative articulation. Here the
into each sentence means extra component delivering procedure remains called Utterance Normalization. Second,
a humoristic impact. Any particular module during there found subtasks from normalization calls before getting
the antinomies discovery (of things, action words, precondition (utterance precondition, current precondition)
modifiers, and descriptive words) in a sentence must during every subtask. That procedure remains called the
be built up; the module misuses the gram-matical opening subtask from call transcripts. Finally, task structure is
connection characterized toward the lexical lexicon extracted it is encoded to AIML and manufacture task-
WordNet. arranged chatbot. The procedure is named as AIML
iii. “Adult slang”: Under numerous situations, some Generation.
funny result is because of the nearness words relating The fundamental offerings are:
the sexual space, for the most part, utilized within 1) Domain-specific tasks work is a combination of related
each adult slang. This chatbot breaks down the pre-conditions and sub-assignments. Pre-conditions are free to
nearness of the sort about the words utilizing rundown arrange to characterizes the substantial orderings like sub-
terms assign adult slang. tasks.
Every humoristic conversational specialist is capable to 2) In human-human discussion to find pre-conditions and
produce humoristic articulations, proposing to the client sub-tasks by Introduce a proficient unsupervised tech-nique.
questions, telling jokes, and amusingly offering an expla- 3) Task-oriented chat-bot are manufacture by encoding sep-
nation to the client. arated task structure with the AIML
Furthermore, the chatbot can recog-nize, during the In utterance normalization, normal dialects give various
discussion with the client, the nearness of humoristic approaches to express something very similar. We have to
articulations, tuning in and making a decision about fun to group and canonicalize the semantically comparative expres-
respond to shaping this visible articulation of the symbol, sions with the goal that are alluded into their accepted
according to these observed levels of entertainment. This chat- structure rather than just a group design.
bot gives a grinning expression if the client’s decision seen as 1) Named Entity Annotation: A standard-based named en-
“funny” if none of the high-lights portraying a humoristic tity annotator that employments many word references and
expression is perceived in the sentence. Most entrancing and rules to explain different sorts of substances in the transcript
complicated views of individual conduct are cleverness. The information. The kinds of elements are LOCATION, DATE,
acknowledgment strategies can be further improved. TIME, AMOUNT.
2) Utterance Clustering: For every utterance, feature vector
C. Automatically extracting dialog models from
are produced. An n-gram is used this shows up an utter-ance
conversation transcripts
being featured. However, utilizing just n-gram fea-tures for
S. Negi et al. [11] exhibited a task-oriented general con- grouping expressions don’t function admirably. For instance,
versation dialog mode that cooperates by the client to achieve analyze each accompanying three expres-sions: (1) do you

0159

Authorized licensed use limited to: University of Edinburgh. Downloaded on September 05,2020 at 02:21:18 UTC from IEEE Xplore. Restrictions apply.
have a substantial credit card, (2) do you have a legitimate The framework shown in Fig. 3 comprises three segments
AAA member card and (3) do you have your credit card that are: client, server, and content acquisition. The server is a
certain utterances should some good number from n-grams like simple ob-ject access protocol (SOAP) mindful web
manner, they semantically not comparative. N-grams utilizing application (web administration) given a black-box approach.
gathered them. The black-box ap-proach secludes the customer from
For task arranged discourse frameworks, the task/domain collaborating with inward operations of the web administration
explicit discourse data, for example, steps in an errand and rather than a white-box approach, inward operations are
area keywords should be given. Indicating this data physically fundamental and permit the customer to interface with a
is a time-consuming procedure. Additionally, the handmade conveyed situation. As showed up in Fig. 4, messages are
information might be brittle and rigid and may not mirror the completely designed in an XML and summarize as a SOAP
client’s view of a task. message pack. Text-based packs are a greater diversity of
clients and platforms. The customer carries voice
acknowledgment handling module which enables the customer
to sent and get plain content.
1) The customer application is effectively open through the
customer browser.
2) The server communications are text and XML formatted.
3) XML messages comply with a pattern that describes the
configuration.
4) Server communications are black-box arranged and parse
Fig. 3. Overview of the Proposed Method [11] approaching XML messages flawlessly.
5) The system allows the client to log in and register into the
D. An intelligent web-based voice chat bot
server and personalized, authenticated and controlled
communication with the server.
6) Content information or voiced sources of info are utilized
by the customer that are effectively comprehend by the
framework.
7) Oneself preparing AI module counteracts an assistance
bottleneck, and accordingly keeps modules from compet-ing
for resources.
The online chat framework follows the customer service
approach that secures the login and runs to the server. Voice
information at that point handled and a reaction is produced.
This process puts a huge preparing necessity on the server’s
processor and memory assets. the server’s processor and
memory resources containing huge procedure. A large number
of clients use the framework at the same time such limitation is
much progressively apparent. Voice acknowledgment needs a
Fig. 4. System Architecture [10] two-section procedure that catches and examination of given
signal input [9]. Every customer uses a working structure input
S. du Preez et al. [10] shows the construction and ad- mechanism to procure a sign, it means to translate the signal to
vancement of an intelligent voice affirmation chatbot. This the customer.
introduces an innovation demonstrator to confirm a proposed
system that helps such a bot (web administration). The com- E. Enterprise crowd computing for human aided chatbots
munication between the client and chatbot through the web- Alessandro Bozzon et al. [1] presented an approach, Human
administration is controlled by the utilization of a black-box Aided Chatbots(HAC) abuse person knowledge, for example
approach, web-administration enables a wide range of by full-time representatives, gaps are filled by restrictions of
customers to convey in any stage of server. XML based totally robotized game plans. Facebook M is a continuous
administration gave is open through a created interface. By instance of HAC. Accomplish more extensive reception, Hu-
presenting an artificial brain, the online-based bot altered man Aided Chatbots must survive various issues, including
client reactions and adjusted the needed character. users ask versatility, low-inactivity, and security. HAC will be chatbots
the inquiries to the bot, which is not comprehended that utilize Crowd Computing. Facebook M1 is an individual
additionally handled utilizing an outsider master framework assistant that arrange Facebook Messenger stages and fruitful
(online intelligent assistant), furthermore, documented reaction case of HAC. Facebook MI permitted custom value-based
improving the artificial brain capacities concerning a gathering assignments with the help of Facebook representatives. Human
of individuals yet to come of reactions. supported bots like Facebook MI assist several clients, for the

0160

Authorized licensed use limited to: University of Edinburgh. Downloaded on September 05,2020 at 02:21:18 UTC from IEEE Xplore. Restrictions apply.
use of HAC, require group processing arrangements that III. CONCLUSION
increase chatbot users’ growth, the expenses related with Integrated Systems performed a significant job within
human computation would become just step by step different fields like customer co-operation. This means hard to
(adaptability), and latency because human communication will each visitor to understand the framework effectively with no
be real. Privacy is another problem as chatbot interactswith the rules. For this situation, extra service is given by the chatbot to
user to share their data, new techniques should be structured the observer. During the article, any incorporated frame-works
and created to address protection concerns. AIML based chatbots are added to their framework to build
collaboration among the client. Different packages and
lightweight AIML files that execute the system extra flexible
and interactive to practice under several fields. Also, a mech-
anized conversational operator based framework performed
some important function to associate beside the client. In this
way, the specialist provides less cost to the automated
conversational agents rather than a personal conversational
agent. The chatbot is performed in the college domain also in
other domains like medical, forensic, sports, etc. Effective for
Fig. 5. System Architecture [1] obtaining information without spending much time.
Fig. 5 shows the reference chatbot design. Given a client REFERENCES
demand, the language understanding (LU) segment deduce the
[1] Alessandro Bozzon, “Enterprise Crowd Computing for Human Aided
customer’s point and related data; the action execution and Chatbots”, ACM/IEEE 1st International Workshop on Software Engi-
information retrieval (AEIR) part plays a mentioned activities, neering for Cognitive Services, 2018.
recovers the data of enthusiasm of information data; the [2] Sarah Bashirieh, Sepideh Mesbah, Judith Redi, Alessandro Bozzon,
Zoltan´ Szlavik,´ and Robert-Jan Sips. 2017. Nudge YourWorkforce: A
response generation (RG) segment manufactures reply into Study on the Effectiveness of Task Notification Strategies in Enterprise
each client; at long last, the dialog management (DM) segment Mobile Crowdsourcing. In Proceedings of the 25th Conference on User
update dialog to missing data and provide clarifications to Modeling, Adaptation and Personalization (UMAP ’17). ACM, New
York, NY, USA, 4–12. https://doi.org/10.1145/3079628.3079692
customers inquiries.
[3] Benjamin Timmermans, Zoltan´ Szlavik,´ Manfred Overmeen, and
Our involvement in-crowd stage proposes Enterprise Crowd Alessandro Bozzon. 2017. ECrowd: Enterprise Crowdsourcing for
Computing that supports HAC as far as to limit and security, Train-ing Cognitive Systems using the Workforce, In Proceedings of the
yet more work is required. For example, it is basic to create 29th Benelux Conference on Artificial Intelligence. BNAIC 2017.
[4] Jean-Franc¸ois Bonnefon, Azim Shariff, and Iyad Rahwan. 2016. The
worker’s commitment techniques ready to adapt the realtime social dilemma of autonomous vehicles. Science 352, 6293 (2016),
necessities of communication specialists. Every one of the 1573–1576.
issues above is responsible for constraints the rule-based algo- [5] Alessandro Bozzon, Marco Brambilla, Stefano Ceri, Andrea Mauri, and
Riccardo Volonterio. 2014. Pattern-Based Specification of Crowd-
rithms, or lacking the arranged model and dataset are sourcing Applications. In Web Engineering, Sven Casteleyn, Gustavo
onesided. Imperatives of automated solutions are completely Rossi, and Marco Winckler (Eds.). Springer International Publishing,
addressed by the use of human intelligence. Human Cham, 218–235.
[6] Alessandro Bozzon, Piero Fraternali, Luca Galli, and Roula Karam.
intelligence could be utilized as constraints to address 2014. Modeling CrowdSourcing Scenarios in Socially-Enabled Human
completely automated so-lutions. Group Computing [5, 6] is a Computation Applications. Journal on Data Semantics 3, 3 (01 Sep
computational worldview that benefactors the use of humans 2014), 169–188. https://doi.org/10.1007/s13740-013-0032-2
dealing with the ability to deal with issues that computers [7] R. Mahapatra, N. Sharma, A. Trivedi, and C. Arnan, “Adding
interactive interlace to e-government systems using aiml based
cannot yet comprehend. chatterbots,” in Soft-ware Engineering (CONSEG), 2012 CSI Sixth
Undertakings receiving crowd computing to continue busi- International Conference on. IEEE, 2012, pp. 1-6.
ness and procedures, demonstrations of publicly supporting are [8] G. Pilato, A. Augello, and S. Gaglio, “A modular architecture for
adaptive chatbots”, Fifth IEEE International Conference on Semantic
moving from online condition to the inner horde of the Computing (lCSC), pp. 177-180, 2011.
venture: representatives. Enterprises Crowd Computing [9] Maja Vukovic and Claudio Bartolini. 2010. Towards a research agenda
(ECrowd [2, 3]) varies from conventional group figuring for enterprise crowdsourcing. In Proceedings of the 3rd Inter-national
Symposium On Leveraging Applications of Formal Methods,
concerning both the group it incorporates (i.e., agents) and the Verification and Validation (ISoLA). Springer, 425–434.
issue it targets (i.e., business issues) [8]. Group Computing [10] S. du Preez, M. Lall, and S. Sinha, “An intelligent web-based voice
gives a compelling method to misuse the inside information chatbot”, IEEE Conference on Advanced computational Intelligence,
pp. 386-391, 2008.
profiles representatives and use their non-used working abil-ity
[11] S. Negi, S. Joshi, A. Chalamalla, and L. V. Subramarnam, “Automat-
to understand business essential basic errands[9]. Also, ically extracting dialog models from conversation transcripts”, Ninth
representatives work under agreements marked to focus on IEEE International Conference on Data Mining, pp.890-895, 2009.
corporate standards and qualities, including, for example, [12] Augello, G. Saccone, S. Gaglio, and G. Pilat, “Humorist bot: Bringing
computational humour in a chat-bot system”, IEEE International
intellectual innovation protection and rights[9]. Responsibility Confer-ence on Complex, Intelligent and Software Intensive Systems
builds representatives constantly accessible for commitment, (ICISIS) , pp.703-708, 2008.
perhaps in an artful way [2].

0161

Authorized licensed use limited to: University of Edinburgh. Downloaded on September 05,2020 at 02:21:18 UTC from IEEE Xplore. Restrictions apply.