See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360887849

An Overview of Chatbots using ML Algorithms in Agricultural Domain

Article in International Journal of Computer Applications · May 2022

DOI: 10.5120/ijca2022922082

CITATION READS

1 336

3 authors, including:

Prabavathy Balasundaram Lekshmi Kalinathan

Sri Sivasubramaniya Nadar College of Engineering VIT University
21 PUBLICATIONS 114 CITATIONS 13 PUBLICATIONS 38 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Prabavathy Balasundaram on 20 March 2023.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

An Overview of Chatbots using ML Algorithms in

Agricultural Domain

Pravinkrishnan K. Prabavathy Balasundaram Lekshmi Kalinathan

Sri Sivasubramaniya Nadar Sri Sivasubramaniya Nadar College Sri Sivasubramaniya Nadar College
College of Engineering, Chennai of Engineering, Chennai of Engineering, Chennai

ABSTRACT identify the intent with which the query was asked. After
The agricultural sector plays a vital part in a country’s which, a similarity function uses the intent and the processed
economic growth. It has already made a major contribution to query to try and get an answer from the existing knowledge
advanced countries’ economic growth. The impact it hason base, that best fits the query. Finally, the extracted answer is
less-developed countries’ economic development is vitally presented back to the user as human-like response.
important. The farmers involved in agricultural activities lack
the resources to stay updated with the information related to
2. LITERATURE SURVEY
the latest advancements in technologies and farming practices. Jain et al., [1] have used a simple Neural network-based
Existing human-involved operations such as Kissan Call model to create an agribot, that can be accessed through any
Center (KCC), even though capable of delivering expected electronic device, to answer queries from farmers. Their
results, has its own drawbacks. Hence there is a need for an model is based on the dataset from Kissan Call Centre which
automated chatbot system that can function as a substitute to consists of previous queries and their answers recorded state
KCC. A chatbot system is a system that delivers domain- wise. The input queries are processed in the neural network
specific knowledge to its users. Such a system in the field of which extracts an entity from it and groups it with similar
agriculture is very helpful in keeping the farmers updated. In sentences using the Sen2Vec Model, after which the most
this paper, existing works on such question-answer systems relevant answer to the query is outputted. Their
focusing entirely on works involving machine learning implementation has improved the accuracy of a sentence
techniqueshave been reviewed. Suggestionsto improve the embedding model from 56% to 86%.
overall usability of the existing systemshave also been made. du Preezet al., [2] have developed a web-based voice chatbot,
implementing a black-box approach. The chatbot takes a
Keywords voice input which is formatted to an XML type and
Chatbot, Query processing, Intent identification, Similarity encapsulated as a SOAP message pack. There on, by the use
function, Answer extraction. of an artificial self-learning brain, the chatbot generates
appropriate responses to the user’s query. Each query and
1. INTRODUCTION responseare archived and used to improve the capability of the
Agriculture is the basis for the sustainability of human life and system. The system uses a modular design for all its
it plays a major part in the development of a country. The components. This distributed environment reduces the overall
developments in bio-technology and chemical engineering has load on one system when compared to systems that run on a
had a huge impact on making agriculture more profitable and single module. The usage of a self-training AI model is
easier. But the shortage of awareness and knowledge claimed to prevent ser-vice bottlenecks and the competing of
regarding these new developments and technologies is still resources by modules.
persistent among conventional rural farmers. On the other end
of the scale, urban farmers who take up farming lack even the Vijayalakshmi et al., [3] have presented a chatbot based
basic ideas behind agriculture. On either side, the problem solution to interact with users and use that to process and
arises due to the shortage of latest and updated knowledge provide solutions to their queries. They have used NLP to
related to specific problems that they face. A chatbot is a parse the input sentences and identify the keywords from
simple system that generates answers to human queries in a them to match them with the existing knowledge base to
particular domain, based on existing knowledge. produce an appropriate answer to the query. The root words in
the query are filtered and converted to a bag of words and
then converted to a vector form so that they can be processed.
A neural network-based model is constructed to classify the
pre-processed data. The neural network was then optimized
using gradient descent to produce the best results on the
Fig 1: Workflow of a chatbot. training data set. They have also incorporated an existing
prediction algorithm, ARIMA, into their system to help
A significant amount of research has been done in the
predict the future cost of agricultural products. This system
betterment of the ability of the chat-bot to generate accurate
takes input through text and voice and can also output its
answers to the given query and also present it in a human-like
answer through text and voice. This paper focuses mainly on
way. Like in a factory as shown in Fig. 1, a chat-bot has an
improving the relevance of the output produced by the
assembly line of processes that it needs to do to generate an
intelligent machine to make it more personal and factual.
answer for the query presented to it. To start with, the query
received is analyzed and processed to convert it to a machine Gawadeet al., [4] have developed an intellectual chatbot
understandable way. Secondly, the processed query is used to system that can respond to queries by students on college-

15
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

related matters. This system has a step-wise regiment to documents that contain keywords equivalent to those present
process and answer each query. First, the chatbot gets the in the query. Based on the documents retrieved by the model,
query as input from the user. Then, the query is pre-processed an answer is extracted from the vocabulary of words. The
to remove stopping words and the remaining keywords are RNN system is claimed to produce answers with high
fetched. Further, the keywords are matched with the existing efficiency.
knowledge base to muster up an appropriate response. Finally,
once the response is returned to the bot, the chatbot packages Sawant et al., [10] have proposed an interactive web-based
the data into a proper response to be displayed for the user. system to assist farmers in agricultural activities. The system
takes various parameters like rainfall, temperature, area and
Yashaswiniet al., [5] developed a smart chatbot based to previous yield into account to answer the queries. The model
answer questions related to agriculture practice and had been trained using crop data from data.gov.in and rainfall
technology. Their model is based on data collected from data from maharani.gov.in both of which are datasets based
various sources like government websites and repositories. on farming and rainfall in Maharashtra. They have trained and
They have used the K Nearest Neighbors (KNN) algorithm as compared the testing and training accuracy of algorithms like
the similarity measure to return the predicted class. The whole KNN, decision tree and random forest. The random forest
system is implemented using a Django architecture, which algorithm is observed to be producing the highest testing
consists of three modules to ensure the proper functioning of accuracy of 78%.
the system. The system updates its knowledge base with each
query that it answers which in turn, improves the accuracy of Lalwaniet al., [11] have proposed an AIML based chatbot to
the system as a whole. This system is claimed to provide answer college-related queries. The proposed system can be
answers with an accuracy of up to 90%. incorporated into any Institution’s website as a pop-up chat
feature. The system AIML files to store question-answer pairs
Jain et al., [6] developed a conversational agent to answer and uses that as the knowledge base to match the query to the
farmer queries that have a simple UI that can be used easily. patterns listed in the existing knowledge base to extract a
The system has been developed to be used in two ways, the response. The query goes through a pre-processing stage
Audio-only FarmChat and Audio + text FarmChat. They have where the keywords are separated using techniques like
developed the model using the KCC dataset and information Lemmatization and POS tagging. The keywords are then
collected from formative interviews with farmers and agri- matched to the knowledge base using Path similarity and Wu-
experts. The proposed system uses cloud-based scalable Palmer similarity tests. The system also maintains a log file
services to implement the conversational agent and a where the inputs to which the system could not produce a
Sequential neural network for training the model for the response are stored. These log files are reviewed and the
chatbot. The speech-based conversational chatbot caters even unanswered queries are answered by an admin and are used to
to the low-literate and digitally illiterate users. The system improve the knowledge base of the system.
has been built based on inputs from potato farmers and rural
India and focuses mainly on the usability and viability of such Feineet al., [12] have covered a variety of areas in which
smart systems among farmers. improvements can be made to existing technologies to create
chatbot systems that generate better responses to human
Ong et al., [7] have created a domain-specific chatbot system queries. They have studied the existing usage of adjectives,
that acts as a platform where farmers can get to know each adverbs and verbs in a sentence by humans and existing
other and share their experiences and easily approach the chatbot dialogues from the ConvAI2 challenge. They have
experts in agriculture to solve their problems and seek proposed chatbot improvement responses mainly along two
recommendations. As an added benefit to agricultural areas, the introduction of a chatbot developer to review and
researchers, they can work together with farmers in helping improve the quality of the generated response, and the
their research and explore more into their research domain. mechanism’s restrictiveness.
The proposed system is a web-based information sharing
platform involving a database that is managed with Python Han et al., [13] have presented a keyword QA system for
using MongoDB. The chatbot is a rule-based one that uses linked data to interpret every possible user intention and
NLP to make the user convenient when interacting with the report related answers in NL form. The system is said to
Chatbot and looking for information. extract every possible triple of information from linked data
that are related to input keywords and report the extracted
Arora et al., [8] have developed a Telegram bot that can information in NL. This system uses entity disambiguation
answer crop disease and weather-related queries. The system and distributed word similarity to match keyword-related
uses a multi-layered CNN based approach which is trained entities and properties in linked data. The query is input in
using the KCC dataset which contains logs of calls at KCC by keyword forms which is used to generate SPARQL queries to
farmers. The weather prediction feature of this system is extract the possible triples. The triples are then used to
implemented by using the OpenWeather-Map API. For the generate reports in natural language.
disease detection feature, the queries are processed and are
con-verted to a word vector which is compared with the Singh et al., [14] have done a detailed survey of the general
existing knowledge base to classify the query to a particular architecture of a chat-bot system and have discussed the
disease. The model is said to produce answers with an improvements made to generic, mainstream chat-bot
accuracy of 70%. assistants. The chatbots they have discussed are ELIZA by
Joseph Weizenbaum, PARRY by Kenneth Colby,
Niranjan et al., [9] have designed a chatbot by implementing a JABBERWACK by Pollo Carpenter, ALICE by Richard
sequence-to-sequence deep learning RNN system to answer Wallace, SMARTERCHILD by Activebuddy Inc., WATSON
farmer queries related to agriculture. The system works in by IBM, SIRI by Apple Inc., MITSUKU by Steve Worsick,
three phases, question analysis, document processing and CORTANA by Microsoft, ALEXA by Amazon, TAY by
answer extraction. The queries are processed in the analysis Microsoft Research, Google Assistant by Google and Bixby
phase through POS tagging, stemming and removal of by Samsung.
keywords. The processing step uses these keywords to fetch

16
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

Liu et al., [15] have improved how chatbots respond to user constructed with ML, applied to the data collected in the RPG
queries by attempting to make those responses more pilot sessions. For the pollution predictor, they have
personalized and human-like. The system uses user data to implemented 4 algorithms, Linear regression, SVR, regression
gather information about their interests, characteristics, tree, and random forest regression, to compare and contrast
language style, etc. The system takes a post and converts it the accuracies produced by each of them. Based on their
into a word vector. After which 3 vectors, post, response and conclusion, the Random Forest Regression algorithm
user, are maintained to classify the post and fill up the produced the least MAE and MSE scores of 0.3948 and
response based on the user. The response with the highest 6.5115 respectively and the highest R2 Score of 0.9907.
confidence score for the given post and user is then displayed.
Patel et al., [22] have discussed the various ways in
Jadhav et al., [16] have developed a crop recommender agriculture in which Machine Learning could be used to help
system that can also give information about weather and farmers. They have also given a general insight into the
locations of fertilizer vendors. The model has been trained various stages in Machine Learning. The applications
using a self-made dataset using the decision tree regression discussed here are the usage of crop selection and crop yield
algorithm to pro-duce the answers to user queries. They have prediction using ML techniques like KNN, ANN, Decision
used Socket programming to implement the system as a Trees, etc., water management in agriculture using regression,
chatbot application and have used APIs to answer weather and and chatbots that answer queries related to farming. This
vendor location queries. The system is claimed to have paper has described various machine learning models and the
produced 90% testing accuracy. scope of machine learning in various applications of the
agriculture field.
Wang et al., [17] have proposed a method to generate answers
according to the in-put question and its similar questions. The Mostacoet al., [23] have developed AgronomoBot which was
proposed method mines the mapping relationship of developed focusing on the search and display of data acquired
questions-answers, questions-questions through a constructed from a Wireless Sensor Network deployed on a vineyard. It
matrix. The answer is generated by making use of the has been developed as a Telegram Bot API and can access
relationship and extracted focus information. They have also information collected by eKo field sensors, bringing it back to
shown experimental results to prove that the proposed method a user through interaction over the Telegram application. The
is promising. IBM Watson cognition services platform was used for
improving the user experience by enabling the use of natural
Wang et al., [18] have proposed a system to substitute tutors language during the conversation experience, providing
in answering student queries in online learning platforms. The intention detection. AgronomoBot is a chatbot that uses NLP
responses are also analyzed based on three metrics, and AI to interact with the user and search for the desired
correctness, professionalism and timeliness. The system takes information in a WSN, adapting to different forms or
in student queries, separates keywords and uses pattern languages of dialogue to achieve the same intention.
matching techniques to find out the best possible response
from the knowledge base. The response is then converted to Fernandes et al., [24] have discussed the implementation and
sentence form using NLP to provide a human-like response. efficiency of various mainstream and domain-based chatbots.
The intelligent teaching assistant system adopts MVC's The domains in which AI and ML-based chatbots were
framework mode, namely model M, view V and con-troller C, implemented that are discussed here are medical, psychiatric
and maintains the intelligent teaching assistant system through counsel-ling, weight control, agriculture and E-commerce.
Django. They have given a generic template of the factors to address
and the process to follow while building such systems. Their
Vijayabaskaret al., [19] have aimed to create a device template suggests places to collect datasets to train the ML
implementing IoT to predict the crop which will yield the Model, ML/DM algorithms that can be used to develop a
maximum profit for that particular soil in that particular model, usage of Dialogflow which gives users ways to
harvest. They have analyzed 5 different implementations of implement interactive conversational systems using NLP,
the predictive algorithm which have considered various soil usage of MATLAB IDE for numerical computations like
properties to predict the crop. They have observed that various matrix manipulations, data plotting, etc., and using AIML for
soil samples taken from different places can be tested using creating conversational agents as it is easy to understand and
the NPK sensors which are portable and have low time highly maintainable.
consumption and prediction based on the atmosphere which
would not be accurate, as the climatic conditions may differ. Chen et al., [25] have done research aimed at building an
intelligence commerce platform system and apply it to the
Yamada et al., [20] have proposed a system to process queries innovative management of agricultural firms and farms. This
through statistical machine translation instead of entity study refers to the interfirm synergy, including the highly
extraction to improve the exactness of the query that is input. uncertain phenomenon of technology and market, and the
The sentence which has the maximum translational formation of clustering synergy. This study also uses the
probability of all the possible translations of the input query is systematic development research methodology as the
computed using the Bayes rule. Using the extracted question development step of this research system. The system uses
sentence and the original question sentence, the ideal answer sequence-to-sequence (seq2seq) architecture to implement
sentence is generated based on the existing learning data. The Machine Translation, Text Summarization and Conversational
system mainly concentrates on Who~?, When~? And Modelling. The intelligence customer service system uses a
Where~? questions and has produced 74%,67% and 80% Neural Conversational Model to generate answers to queries.
accuracies respectively and overall accuracy of 74%. The research results combine ICT and intelligence on the e-
Alves et al., [21] have done work to present the development commerce platform to develop intelligence business with
of a chatbot capable of assisting different RPG roles in exemplary and standard, apply GPS positioning system to the
decision-making. Game agents consult trends using statistics farm and develop intelligence customer service system with
and make predictions about pollution levels based on a model deep learning calculus.

17
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

Nayaket al., [26] have developed a chatbot which is designed authorities. The evaluation parameters used are Mean
to act as a farming assistant that clears all the doubts of the Absolute Error and Root Mean Squared Error. The paper
farmers in an efficient manner. The proposed system uses concludes that the Decision tree algorithm produces the best
Machine learning algorithms to generate query responses and result for the given dataset with an accuracy of 98.62%, MAE
the system fails to respond to the farmer’s queries, these of 1.45 and RSME of 2.11 compared to other algorithms like
queries are forwarded to experts. The knowledge base for the Linear Regression (89.38%), Lasso Regression (86.33%) and
chatbot is obtained from the data that is collected and grouped Ridge Regression (89.53%).
from the most asked question and answers about farming or
agriculture from the internet. The proposed chatbot uses the Manjula et al., [29] have built an architecture for crop yield
Levenshtein distance formula for calculating the difference in prediction besides pro-posing a new methodology that
the string and returns a value which is named as the combines the usage of vegetation indices derived from remote
confidence value. The system they have proposed is claimed sensing images and other attributes. The framework is
to answer queries with an accuracy of 96%. designed to be flexible and dynamic so that it can be used for
crop yield prediction for different crops. Their methodology
Kale et al., [27] have done research proposing the Neural makes use of vegetations indices collected through remote
Network model to predict crop yield and success rate of crops sensing technology, climate-related variables, agronomic
depending on the dataset provided by the Indian government. related variables, and weather disturbance information. The
The dataset is huge containing data for all the regions of India proposed approach uses the crop yield model and data mining
which were filtered to get data for Maharashtra state. The crop approach coupled with the crop yield model for accurate
yield prediction model uses the backpropagation algorithm of prediction results.
an Artificial neural network. A multilayer perceptron
technique is used. The performance of the model is evaluated Momayaet al., [30] have proposed a system that answers
using parameters like Mean Absolute Error, Mean Squared queries related to weather, plant protection, animal husbandry,
Error and Root Mean Squared Error. The ANN with linear market price, fertilizer uses, government schemes, soil testing,
regression with forward and backward propagation model etc. The data for the database is collected from the official
predicted the dependent variable with 82% accuracy and very website of Kisan Call Centre (KCC). The system takes the
little loss. input query and gives it as an input to RASA NLU which then
classifies it into intents, identifies entities and finds out the
Kavita et al., [28] have done research estimates the crop yield matching intent with the highest confidence score. Further, it
for India using data from 1950 to 2018. The prediction is is checked if the intent corresponds to the weather query or
made for five crops which are Rice, Wheat, Jowar, Bajra, not. If the intent is not a weather query, then its corresponding
Tobacco, and Maize using parameters including the area used intent is found from the database. The answer to the query
for the crop sowing, production, Yield, and Area under from Rasa core is then channeled to ngROK to Twilio and
irrigation. The prediction is attained using Decision tree and finally to WhatsApp for easy access by the user. The system is
Random forest. The dataset used for the experiment in this claimed to have 95.67% Intent Accuracy, 94.285% Story
research was collected from ww.mospi.gov.in and Accuracy, F1 score of 97.90 and overall Precision and
https://data.gov.in, which is made public by government Accuracy of 98.46% and 96.1% respectively.

Table 1. Analysis of the various chatbot implementation strategies and their limitations.
S.No References Methodology Reported testing accuracy Limitations

1 [1], [2], [4], [5], Chat bots were developed [1]-86%, [5]- 90%, [10]-78%, • Needs proper internet connectivity
[7], [10], [25], aswebsites for the end users [26]-96% to be accessed
[26] whoare farmers • Does not cater well to digitally
illiterate people

2 [6], [11], [19], Chat bots were developed to [31]-75.55% • Huge amount of storage is required
[21], [31], [32] workin standalone machines asthe whole system is stored on the
withsimple UI user’s device
• Updates and bug fixes require data
connectivity and high amount of
download able data

3 [3], [8], [23], [30] Chat bots were built as [8]-70%, [30]-intent-98.46% • Failure in API connectivity results
Telegramand Whatsapp story-96.1% in failure of the whole system
using appropriate APIs
• Does not cater well to digitally
illiterate people

4 [1], [2], [4], [9], Answers extracted from the [1]-86%, [9]-78%, [16]-90%, • Keywords in their raw form are
[10], [16], [19], chatbots were keywords [28]-98.62%, [31]-75.55% difficult to interpret
[22], [27], [28], • Misinterpreted keywords might
[29], [31] misguide the users

18
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

5 [3], [5], [8], [11], Chat bots generates answers [5]- 90%, [8]-70%, [20]-74%, • Lack of native language support.
[13], [18], [20], whichthe human can easily [21]- R2 Score of 0.9907,
[21], [25], [26], understand [26]-96%, [30]- intent 98.46%
and story 96.1%
[30], [32]

6 [6], [17], [23] Chat bots generates answers Not reported. • Language support is limited tovery
whichthe human can easily few popular languages
understandwith the language
selected by theuser

Medaret al., [31] have presented a comprehensive way in Geographic Information System (GIS) techniques to monitor
which machine learning algorithms can make use of the crop yield and its variation rate. The research presented in
agricultural factors to predict the yield of a particular crop in a this paper has focused on the use of fusion techniques to solve
particular field. They have implemented and compared the the problem of crop yield modelling, including measurement
performance of Machine learning algorithms like Naïve Bayes to statistics result association, track fusion, and decision
and KNN in predicting the crop yield. They have concluded fusion in complex environments of remote sensing. A fusion
that, for the dataset they have used, the Naïve Bayes algorithm technique for measurement to statistics result association has
produced a better testing accuracy of 91.11% compared to the been proposed. This approach is based on the data fusion
KNN algorithm, which produced 75.55%. model and is suitable for TM and statistics sensors having
different types of attributes. An optimum fusion method for
Gounderet al., [32] have developed a mobile application that combining the output of TM and statistics sensors in the
will help farmers by answering agriculture-related queries. system of crop yield modelling has also been investigated.
The dataset used to develop this system has been obtained
from data.gov.in that consists of data from recorded calls from 3. ANALYSIS OF EXISTING WORK
KCC Odisha. They have used the NLTK library to implement This section discusses the existing literature with respect to
NLP algorithms for extracting the required information to be key features, namely, implementation, query analysis, intent
converted to a word vector. The word vector is used to identification, similarity function and answer presentation.
compare to the existing knowledge base using a cosine
similarity function, which outputs an appropriate answer to 3.1 Implementation
the user query. A farm help system, even though being a simple idea to
Palasundramet al., [33] have worked to improve the sequence- comprehend, the implementation of such a system has several
to-sequence (Seq2Seq) model for natural answer generation in factors that determines its usability. One of the major factors
chatbots. This literature review has identified and reviewed that needs to be considered in the development of such
the methods proposed to address the weakness such as systems to be used in agricultural domain is accessibility to
utilizing additional embedding and encoders, using different the target audience. Out of the various ways, messenger-based
loss functions and training approaches, as well as utilizing Chat/Talk Bots, web based conversational agents and
other mechanisms like copying source words and paying dedicated mobile applications are observed to be the most
attention to a certain portion of the input. They have proposed usable and accessible methods of implementations.
changes such as structural modifications, augmented learning, Messenger-based Chat/Talk Bots:
beam search and complementary mechanisms to the Seq2Seq
model to make it perform better at Natural Answer These are systems that are developed to be incorporated with
Generation. The structural modifications suggested are to existing popularly used messaging applications like Telegram
have additional embeddings and encoders. Augmented [8][23] and WhatsApp [30] using appropriate APIs. Its
learning introduces the usage of approaches like Alternative implementation is such that, the system is easily accessible to
loss function learning, Multi-Task Learning, Deep be used in the same way the user would message any of their
reinforcement learning and Adversarial learning to further contacts in the messaging application.
improve the answer generation quality of Seq2Seq. They have
proposed the usage of the Diverse Beam Search algorithm as Web-based Chat/Talk Bots:
an improvement on the existing Beam Search method. They
have concluded that these enhancements provide support for In this approach the system is hosted on the internet as a
the Seq2Seq model during training and prediction to generate website [5][11][30] to be used by its beneficiaries. The aspect
meaningful answers. of not having to store any files except for cookies in the user’s
smart device, to use the system is a huge advantage as these
Mishra et al., [34] have made an extensive survey of Machine
systems can require a huge amount of storage space.
learning techniques used in crop yield prediction models by
various scholars. The techniques discussed are Artificial
Dedicated Applications:
Neural Networks, Information Fuzzy Network, Decision Tree,
Regression Analysis, Clustering, Bayesian Belief Network,
In this method, an application having a simple UI [6] and the
Principal Component Analysis, Time Series Analysis and
ability to work offline is developed to ensure easy access by
Markov Chain Model. They have given a clear-cut conclusion
everyone. The only drawback in this approach is that all the
on which algorithms produce the best result for which
data required for the application to work has to be stored in
application using a table that shows the algorithm type and its
the user’s device and the response time depends totally on the
application area.
user’s hardware capabilities.
Sun et al., [35] have used Remote Sensing (RS) and

19
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

3.2 Query Analysis were concluded to be the increasing order of their R2 scores
Analyzing the input query and extracting the important [21].
information is the primary step that a chatbot system does.
This step takes care of processing the human language input
3.5 Answer Presentation
by removing stopping words, marking nouns, etc. and This is the final stage in the working of a chatbot, which is
assigning weightage of each word in the processed input, so responsible for presenting the answer extracted from the
that the similarity function can produce an appropriate model in a human-like way. One method of approach was to
response. The various ways in which researchers have include human domain experts [26] in the process to improve
adopted to process input queries are, Sen2Vec [1] which is a the response generation process of a chatbot system. Other
convenient solution to transform sentences to vectors that implementations that do not involve human work include the
contain high level of information of the original sentences. use of RASA X which is an open-source tool for
Usage of SOAP message pack [2] to present the extracted Conversation-Driven Development, template-based Natural
information based on the voice input which is primarily Language Generation, APIs provided IBM cloud [23] and by
converted to an XML type format [2]. Implementation of Bag incorporating neural conversational models [30].
of Words approach by the use of NLP to parse sentences to
extract keywords and root words from the input [3]. Further, a 4. RESULTS OF EXISTING WORK
combination of several task-specific algorithms like For the website-based implementation of chatbots, the testing
tokenization, noise removal, lexicon normalization, stopping accuracies of the models presented range from78% [10] to as
word removal [4], POS tagging [9][11], stemming [9], high as 96% [26]. Out of the few standalone applications with
lemmatization [11], etc., have also been used to develop such simple UI, the reported accuracies were sparse and was found
systems. to be 75.55% [31]. The messenger integrated applications
have reported accuracies ranging from 70% [8] to 96.1% [30]
3.3 Intent Identification for answer generation and with the highest accuracy of
Intent identification is the process of probing into the output 98.46% [30] for intent identification. Among the chatbots that
from the first step to identify and mark the words which form use NLP to produce human-like answers, the accuracy ranges
the basis of the query. Through this process, even after getting from 70% [8] to 96.1% [30].
a word vector from the input, distinct words in the vector that
decide the intent behind the question are given more 5. RESEARCH CHALLENGES
weightage before sending the vector to the next stage in the Kissan Call Center (KCC) is a government organization
process. The techniques using which this process is done which was formed to answer farmer’s queries on a telephone
include tagging nouns in the word vector as important words call in their own dialect. KCC is a really successful initiative
and increasing their weightage, as nouns are the intent that supports farmers by answering their queries related to
deciders in most sentences. Using DBpedia to create entity- farming and many other areas like animal husbandry, Nutrient
property-entity groups which can be queried on using management, Livestock products processing and packaging,
SPARQL [13]. Other implementations include usage of etc. But it takes a lot of money to set up and run these KCC
Statistical Machine Translation [20] to identify words which call centers. Hence, in recent times, a notable amount of
decide the intent based on the frequency of the usage of the research has gone into developing chatbot systems that can
word and RASA NLU which is an open-source conversational replace these call centers. These systems that are currently in
AI that performs intent classification, entity extraction and use can be made to work better if the following ideas are
response retrieval [30]. incorporated into them.

3.4 Similarity Function 5.1 Usage of live information

The third and most important process in the working of a The queries from farmers can be classified into two major
chatbot system is the answer extraction stage. This is where categories, queries based on recorded data and queries based
the pre-processed queries from the previous steps are used to on live data. For example,
extract an answer that is most appropriate based on its
General query: How long does it take to grow paddy in a
similarity to the existing knowledge base. The best fit for a
field?
given query can be obtained by the usage of a similarity
function which has various ways in which it can be Live query: Is it profitable to grow paddy in the current
implemented. The word2vec algorithm uses a neural network climate?
model to learn word associations from a large corpus of text.
It can cluster similar sentences without taking into account of KCC is observed to give back better responses to both kind of
order of the words. Another implementation is a black box queries when compared to automated systems. It is because
approach [2] which implements a self-learning AI algorithm the people who work there have live access to constantly
that find paths to the best answer from the given pre- changing data and are able to make informed decisions based
processed query. Further, classification algorithms optimized on it. However, chatbot systems work based on previously
using gradient decent are also being used as a function to recorded data with which it was trained. The information that
extract answers to user queries. Other implementations was relevant when the data was recorded may or may not be
include K-Nearest Neighbor algorithm [5], a multi-layered relevant when the query is being asked. To address this
CNN approach for image-based queries [8], sequence to problem, the system should take into account live information
sequence deep learning RNN for speech-based queries [25] that are relevant to the query asked and make decisions based
and other algorithms like Path similarity and Wu-Palmer on that.
similarity functions [11]. Further, an analysis was done to find
the efficiency of majorly used ML algorithms. Using the R2
5.2 Location based dataset
Natural features like soil fertility, type of soil, climatic trends,
score as the judging parameter, linear regression, support
availability of water, etc. are major factors that influence
vector regression, decision tree and random forest regression
farming in a particular location. These factors lead us to

20
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

consider that generic processing of queries without Issue 5, (2019)

considering the location from which it was asked leads to the
production of faulty or irrelevant answers. Data collected [6] Jain, M., Kumar, P., Bhansali, I., Liao, QV., Truong, K.,
from one location might not produce appropriate answers for Patel, S.: FarmChat: a conversational agent to answer
queries from other locations. To resolve this, it is suggested farmer queries. In: Proceedings of the ACM on
that the data set be stored as clusters based on the location it Interactive, Mobile, Wearable and Ubiquitous
was collected from. When the system gets an input query, the Technologies, pp.1-22. Association for Computing
location from which the query was asked should be Machinery, New York, NY, United States (Dec 2018).
considered first. The answer should be extracted based on data [7] Ong, RJ., Raof, RA., Sudin, S., Choong, KY.: A Review
collected from the location that the query was asked. This of Chatbot development for Dy-namic Web-based
approach will make the answer extracted to be more relevant Knowledge Management System (KMS) in Small Scale
and appropriate. Agriculture. In: Journal of Physics: Conference Series,
Vol. 1755, Issue 1, p. 012051. IOP Publishing (Feb
5.3 Implementation based on target user 2021).
A chatbot system is commonly developed as a messenger-
[8] Arora, B., Chaudhary, DS., Satsangi, M., Yadav, M.,
based, a web-based or as an app-based. These three
Singh, L., Sudhish, PS.: Agribot: a natural language
implementations should be adjusted to better suit the target
generative neural networks engine for agricultural
users who are broadly split as digitally literate naïve farmers
applications. In: 2020 In-ternational Conference on
and digitally illiterate expert farmers. The messenger-based
Contemporary Computing and Applications (IC3A), pp.
and web-based implementations can be easily used by the
28-33. IEEE, Lucknow, India (Feb 2020).
digitally literate naïve farmers. This is because they do not
have issues related to connectivity and are assumed to be [9] Niranjan, PY., Rajpurohit, VS., Malgi, R.: A survey on
digitally literate. The app-based system updates live chat-bot system for agriculture do-main. In: 2019 1st
information at regular intervals. It can also work without any International Conference on Advances in Information
connectivity. Hence it will be useful for the digitally illiterate Technology (ICAIT), pp. 99-103. IEEE, Chikmagalur,
expert farmers. India (Jul 2019).
[10] Sawant, D., Jaiswal, A., Singh, J., Shah, P.: AgriBot-An
6. CONCLUSION intelligent interactive interface to assist farmers in
Developments in technology and modern systems like ML- agricultural activities. In: 2019 IEEE Bombay Section
based chatbots have enabled farmers to be updated with the Signature Conference (IBSSC), pp. 1-6. IEEE, Bombay,
latest developments in the field of agriculture. There are a lot India (Jul 2019).
of such chatbot question-answering systems, but they fail to
produce contemporary responses. This work was done to [11] Lalwani, T., Bhalotia, S., Pal, A., Rathod, V., Bisen, S.:
analyze the existing research work on chatbots in the Implementation of a Chatbot Sys-tem using AI and NLP.
agricultural domain and suggest how it could be made to International Journal of Innovative Research in
produce better results by taking new factors like live Computer Science & Technology (IJIRCST) Vol. 6,
information, location-based datasets and implementation Issue 3. (May 2018).
styles into consideration. The existing systems can be made to
[12] Feine, J., Morana, S., Maedche, A.: A chatbot response
produce more accurate and relevant answers to farmer queries
generation system. In: Proceedings of the Conference on
by incorporating the suggested changes to the existing system.
Mensch und Computer, pp. 333-341. ACM, New York,
These suggested improvements will help in enhancing the
United States (Sep 2020).
overall usability of such systems. The future scope of research
in this domain lies in improving and extending native [13] Han, S., Shim, H., Kim, B., Park, S., Ryu, S., Lee, GG.:
language support provided by these applications so that it can Keyword question answering system with report
be put to use by a greater number of people. generation for linked data. In: 2015 International
Conference on Big Data and Smart Computing
7. REFERENCES (BIGCOMP), pp. 23-26. IEEE, Jeju, South Korea (Feb
[1] Jain, N., Jain, P., Kayal, P., Sahit, J., Pachpande, S., 2015).
Choudhari, J.: AgriBot: agriculture-specific question [14] Singh, S., Thakur, HK.: Survey of Various AI Chatbots
answer system. IndiaRxiv (June 2019). Based on Technology Used. In: 2020 8th International
[2] du Preez, SJ., Lall, M., Sinha, S.: An intelligent web- Conference on Reliability, Infocom Technologies and
based voice chat bot. In: IEEE EUROCON 2009, pp. Optimization (Trends and Future Directions) (ICRITO),
386-391. IEEE 18 May 2009. pp. 1074-1079. IEEE, Noida, New Delhi, India (Jun
2020).
[3] Vijayalakshmi, J., Pandimeena, K.: Agriculture talkbot
using AI. International Journal of Recent Technology [15] Liu, B., Xu, Z., Sun, C., Wang, B., Wang, X., Wong,
and Engineering (IJRTE), Vol. 8, pp 186-90 (Jul 2019). DF., Zhang, M.: Content-oriented us-er modeling for
personalized response ranking in chatbots. IEEE/ACM
[4] Gawade, H., Patil, V., Vishe, P., Kolpe, S.: College Transactions on Au-dio, Speech, and Language
Enquiry Chat-Bot System. International Journal of Processing. pp. 122-33. (Oct 2017).
Engineering Research & Technology (IJERT), Vol. 9,
Issue 9 (Sept 2020). [16] Jadhav, M., Kolambe, N., Jain, S., Chaudhari, S.:
Farming Made Easy using Machine Learning. In: 2021
[5] Yashaswini, D., Hemalatha, K., Nivedhitha, G.: Smart 2nd International Conference for Emerging Technology
Chatbot for Agriculture. International Journal of (INCET), pp. 1-5. IEEE, Belgaum, India (May 2021).
Engineering Science and Computing (IJESC) Vol. 9,
[17] Wang, R., Ning, C., Lu, B., Huang, X.: Research on

21
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.11, May 2022

answer generation method based on focus information Management and e-Learning, Vol. 10, Is-sue 2, pp. 114-
extraction. In: 2012 IEEE International Conference on 24 (2020).
Computer Science and Automation Engineering (CSAE),
Vol. 2, pp. 724-728. IEEE, Zhangjiajie, China (May [26] Nayak, V., Sowmya, NH.: Agroxpert-Farmer assistant.
2012). Global Transitions Proceedings, Vol. 2, Issue 2, pp. 506-
12 (Nov 2021).
[18] Wang, Y., Sun, Y., Chen, Y.: Design and Research of
Intelligent Tutor System Based on Natural Language [27] Kale, SS., Patil, PS.: A Machine Learning Approach to
Processing. In: 2019 IEEE International Conference on Predict Crop Yield and Success Rate. In: 2019 IEEE
Computer Science and Educational Informatization Pune Section International Conference (PuneCon), pp. 1-
(CSEI), pp. 33-36. IEEE, Kunming, China (Aug 2019). 5. IEEE, Pu-ne, India (Dec 2019).

[19] Vijayabaskar, PS., Sreemathi, R., Keertanaa, E.: Crop [28] Kavita, M., Mathur, P.: Crop Yield Estimation in India
prediction using predictive analytics. In: 2017 Using Machine Learning. In: 2020 IEEE 5th
International Conference on Computation of Power, International Conference on Computing Communication
Energy Information and Communication (ICCPEIC), pp. and Automation (ICCCA), pp. 220-224. IEEE, Noida,
370-373. IEEE, Tamil Nadu, India (Mar 2017). New Delhi, India (Oct 2020).

[20] Yamada, T., Arakawa, T.: A study on output sentence [29] Manjula, A., Narsimha, G.: XCYPF: A flexible and
generation method for question answering using extensible framework for agricultural Crop Yield
statistical machine translation. In: 2013 13th Prediction. In: 2015 IEEE 9th International Conference
International Conference on Control, Automation and on Intelligent Systems and Control (ISCO), pp. 1-5.
Systems (ICCAS 2013), pp. 1199-1202. IEEE, Gwangju, IEEE, Coimbatore, Tamil Nadu, India (Jan 2015).
South Korea (Oct 2013). [30] Momaya, M., Khanna, A., Sadavarte, J., Sankhe, M.:
[21] Alves, BC., de Freitas, LA., de Aguiar, MS.: Chatbot as Krushi–The Farmer Chatbot. In: 2021 International
support to decision-making in the context of natural Conference on Communication information and
resource management. In: Anais do XII Workshop de Computing Technology (ICCICT), pp. 1-6. IEEE,
Com-putaçãoAplicada à Gestão do MeioAmbiente e Mumbai, India (Jun 2021).
RecursosNaturais, pp. 29-38. SBC, Brazil, South [31] Medar, R., Rajpurohit, VS., Shweta, S.: Crop yield
America (Jul 2021). prediction using machine learning techniques. In: 2019
[22] Patel, D., Suryakantbhai, N.: AGRICULTURE WITH IEEE 5th International Conference for Convergence in
MACHINE LEARNING. European Journal of Molecular Technology (I2CT), pp. 1-5. IEEE, Mumbai, India (Mar
& Clinical Medicine. Vol. 7, Issue 8, pp. 5491-5493 (Feb 2019).
2021). [32] Gounder, S., Patil, M., Rokade, V., More, N.: Agrobot:
[23] Mostaco, GM., De Souza, IR., Campos, LB., Cugnasca, An Agricultural Advancement to Enable Smart Farm
CE.: AgronomoBot: a smart answering Chatbot applied Services Using NLP. Journal of Emerging Technologies
to agricultural sensor networks. In: 14th international and Innovative Research (2021).
conference on precision agriculture, Vol. 24, pp. 1-13. [33] Palasundram, K., Sharef, NM., Kasmiran, KA., Azman,
Montreal, Quebec, Canada (Jun 2018). A.: Enhancements to the Sequence-to-Sequence-Based
[24] Fernandes, S., Gawas, R., Alvares, P., Femandes, M., Natural Answer Generation Models. IEEE Access. pp.
Kale, D., Aswale, S.: Survey on var-ious conversational 45738- 45752 (Mar 2020).
systems. In: 2020 International Conference on Emerging [34] Mishra, S., Mishra, D., Santra, GH.: Applications of
Trends in In-formation Technology and Engineering (ic- machine learning techniques in agricultural crop
ETITE), pp. 1-8.IEEE, Vellore, Tamil Nadu, In-dia(Feb production: a review paper. Indian Journal of Science
2020). and Technology, Vol. 9, Is-sue 38, pp. 1-4 (Oct 2016).
[25] Chen, AP., Chansilp, K., Kerdprasop, K., [35] Sun, L., Zhu, Z.: A RS/GIS-Based System for
Chuaybamroong, P., Kerdprasop, N., Shun-Fa, H., Yi-In, Monitoring Crop Yield. In: 2011 International
L., Ching-Tzu, H., Chih-Hua, C., Chen, ST., Zhang, RC.: Conference on Computer Distributed Control and
B2B Marketing Crafts Intelligence Commerce: How a Intelligent Environmental Monitoring, pp. 17-20. IEEE,
Chatbot Is Designed for the Taiwan Agriculture Service. Changsha, China (Feb 2011).
In-ternational Journal of e-Education, e-Business, e-

IJCATM : www.ijcaonline.org 22

View publication stats