2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS)

A Review of Current Trends in the Development of

Chatbot Systems
Tatwadarshi P. Nagarhalli Vinod Vaze N. K. Rana
PhD Research Scholar, Dept. of Computer Sci. & Principal
Comp. Engg., Shri JJT Engg. Shri JJT University Theem College of Engineering
University Rajastan, India Rajastan, India Mumbai, India
[email protected] [email protected] [email protected]

Abstract— The main aim of any technological advancement is customer experience. As a matter for fact it has been observed
to lend a helping hand in making lives of humans easy. This is that consumers prefer chatbot systems over human interaction
also true for the field of natural language processing. This is also and going ahead would prefer these systems over physically
the reason why conversational systems, also called as chatbot visiting the stores [6].
systems have gained popularity in the resent times. Chatbot
system have been adapted and developed for many different These chatbot systems are given knowledge, with the help
domains. The paper conducts a detailed study of some of the of which these systems try and answer user’s queries. This
recent chatbot systems/papers developed in different domains. knowledge given to the chatbot systems are of three types
These recent papers have been reviewed while keeping special structured database, knowledge bases and unstructured dataset
attention to the type of knowledge given to these systems, the [7].
domain for which these systems have been developed, among
other parameters in order to understand the recent trends in the The main focus of the paper is to study the recent trends in
development of chatbot systems. the development of the chatbot systems with respect to the
different domains for which these systems have been
Keywords— Chatbot, Chatbot Systems, Dialogue Systems, developed and the type of knowledge given to these systems.
Conversation Systems, Natural Language Processing, Machine
Learning, Chatbot Knowledge, Chatbot Domain. II. REVIEW OF CHATBOT SYSTEMS
Critical reviews of recent papers on chatbot development
I. INTRODUCTION have been carried out in this section.
The increase in the computing power has paved the way
for new technological advancements. Artificial intelligence The heart of a chatbot system is the knowledge which has
has played a very crucial role in these technological been provided to it. Using this knowledge the chatbot carries
advancements. One of the important application of artificial out conversations and answers the questions raised by the
intelligence is Natural Language Processing. Natural users. But, Sameera A. Abdul-Kader and Dr. John Woods [8]
Language Processing is a method of making the machine or a propose to develop a new chatbot system which does not have
computer understand the human language [1]. And, one of the a knowledge. Rather, the authors propose to make use of the
important application of Natural Language Processing is World Wide Web in order to search for the answers for the
Conversational Agents, also called as Conversational Artificial queries raised by the users.
Intelligent Bots or it can be simply called as Chatbot Systems. The paper proposes to make use of chatterbot and natural
In the recent times chatbot systems have gained popularity language processing toolkit for performing basic natural
because of its wide applications. Apart from the wide range of language processing operations. The paper proposes to
applications the reason for the popularity of chatbot systems is perform text matching for finding the answers for the queries
that they are approachable, they enhance customer experience, and these answers are stored in a structured database.
they can manage large number of customers, and are very cost On the other hand the problem identified by Che-Hao Lee
effective [2]. It has been observed that chatbot systems helps et. al. [9] is that for the development of companion robot for
in reducing the overall operating costs as well [3]. kids it is important to have good question answer pair. The
In the coming years it is expected that chatbot systems will paper proposes to generate questions from about 100 student’s
reduce the workload at the higher management levels by up to tales. The paper makes use of logistic regression for the
70% [4]. Because of this corporations are expected to invest purpose of identifying the answer for the query asked by the
billions of dollars in the research and development of chatbot user. This is done by ranking the question answer pair by
systems [5]. making the use of logistic regression. The details of natural
language processing and the knowledge type has not been
Even from the customer’s pot of view chatbot systems elaborated in the paper.
present a unique experience of availability of help and support
for a product anytime of the day. This help the overall

978-1-7281-5197-7/20/$31.00 ©2020 IEEE 706

 2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS)
Taking this idea forward Ly Pichponreay et. al. [10] develop a machine learning system which will monitor the
propose to use Apache PDFBOX for optical character social media sites for any grievances raised by customers.
recognition of text present in the portable document format.
If the machine learning system finds any actionable
After the text has been identified the paper proposes to
complaints then it will invoke the chatbot system module. This
generate questions using different techniques and then using
chatbot system will interact with the customers in order to find
string matching identify the answers to given queries. The
the specific problems faced by them. The chatbot system will
paper is quite very simplistic in nature. It makes use of very
also try to give certain solutions to the problems. If the
basic natural language processing for completing the required
problems persist and the customer want to file an official
tasks, whereas it is understood that they are much better
complaint then the chatbot system will assist in that as well.
techniques which can enhance the quality of the conservations.
While the idea of scanning social media sites for
The main premise identified by authors Ming-Hsiang Su
complaints is a unique one, the idea of using chatbot system
et. al. [11] is that one of the important demography which has
for customer service is a tried and tested method. Also, the
been conveniently ignored are the elders. Due to many
paper does not dwells into the details of the development of
compulsions is not possible to sit with the elderly and have a
chatbot systems or the capabilities of the chatbot systems.
good conservation. So, the authors propose to build a chatbot
system which engages the elderly people in small talk and Bhavika R. Ranoliya, et. al. [15] on the other hand propose
conservation. For this the paper proposes to create a database to develop a chatbot system in order to answer the frequently
containing more than two thousand message and its response asked question about a particular university. The paper
pairs. The paper proposes to use long short term memory for proposes to make use of Artificial Intelligence Markup
the purpose of maintaining the state information. The paper Language. The paper discusses different tags the language
does not talk about dealing with unstructured data. offers and their use in the development of chatbot systems.
The paper also gives the detailed comparison with earlier two
The problem identified by the authors Shih-Hung Wu et.
methods of the development of chatbot system for a university
al. [12] is that it is quite easy for a robot to tell a long story,
and claims that the use of artificial intelligence markup
but it is difficult for the robot to have a conservation on the
language has many advantages over these systems.
story narrated by it. For this purpose the paper proposes to
develop a template by using CKIP toolkit for word The main drawback of the paper is that the paper very
segmentation and parts of speech tagging. The paper also elaborately explains the different features and tags in the
proposes to use Term Frequency – Inverse Document artificial intelligence markup language, but the development
Frequency (TF-IDF) for noun identification. of chatbot has not been explained in the paper. The type of
knowledge used in the chatbot, the features and accuracy of
The paper makes use of artificial intelligence mark-up
the chatbot has not been elaborated in the paper.
language for the development of the chatbot. The major
drawback of the proposed system is that it makes use for Divya Madhu, et. al. [16] proposes to develop a chatbot
legacy techniques for the development of the chatbot, there are system for the medical domain. The proposed chatbot system
better techniques for the development of a robot which can takes in the symptoms from the users and predicts the possible
carry out conservations on a story. diseases. The paper also proposes to give the medicine and
dosage suggesting capability to the chatbot system.
The wide range of applications of natural language
processing has enabled the use of chatbot systems in many Even though the idea of medical care is useful there are
different domains as well. Authors Cyril Joe Baby, et. al. [13] major drawbacks in this paper. The paper provides the pseudo
propose to use chatbot system in for home automation and code for the disease prediction. From this pseudo code it can
Internet of Things domain. The paper proposes to use natural be understood that each and every disease and symptoms have
language toolkit for the basic natural language processing been stated and the system has been coded very explicitly,
tasks. The main idea is to identify the key words entered by which is a very difficult task and this makes the system non
the user into the chatbot system and these key words are scalable. Also, the implementation of the chatbot system has
passed on to the microcontroller which will control the home not been detailed in the paper.
appliances through Internet of Things.
The main problem identified by Hanjong Choi and
The paper is a very good application of chatbot systems. Takeshi Hamanaka [17] is that when a user buys a new
But, apart from the basic processing operations on natural product, especially any electronic gadgets, he or she has to go
language and identification of keywords the paper does not through the complete manual in order to understand the
dwell much further in the field of natural language processing. working and the features of the product, which is expecting
Also, the performance of the chatbot system has not be the user work in an unknown environment. Also, the
discussed in the paper. understanding of different people differs, some might be slow
to pick up the details while others might be quick. To cater to
On the other hand Godson Michael D’silva, et. al. [14]
this problem the paper proposes to develop a chatbot system.
propose to use chatbot for customer service. The problem
identified by the paper is that the grievances raised on the The idea of catering to the basic and advance users of an
social media by customers are not always acted upon by the electronic gadgets through chatbot system is a unique idea.
product companies quickly enough. So, the paper proposes to But, the paper does not elaborate the implementation of
chatbot system, nor does it talk about the natural language

978-1-7281-5197-7/20/$31.00 ©2020 IEEE 707

 2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS)
processing tasks employed by the paper. Also, the paper does language processing, use of machine learning algorithms, etc.
not talk about the type of knowledge given to the chatbot have not been elaborated.
system.
Chaitrali S. Kulkarni, et. al. [22] propose to develop a
On the other hand Wei Liu, et. al. [18] propose to develop chatbot system for the banking domain. The problem
a completely new chatbot system on the lines of Siri, Alexa or identified by the paper is that when it comes to money people
Google Assistant. The paper says that IBM Watson makes use are, justifiably, extra cautious. Banking services users might
of fragments of knowledge extracted from the query and tries have many questions at any point in the day. But, the
to answer the query rather than understanding the complete availability of customer service representative is not
meaning of the query. guaranteed. So, the paper proposes to develop a chatbot
system for answering customer’s queries.
In order to understand the complete meaning of the query
the paper proposes to make use of DeepQA system. The paper The paper proposes to develop a dataset by downloading
proposes to use the internet for finding the answers to the the frequently asked questions section of the different banking
questions asked by the users, this makes the system open platforms. The paper proposes to employ natural language
ended. The paper also discusses the usefulness of the system toolkit for the processing of natural language. The paper also
proposed. But, the technologies used for the development of proposes to use bag of words method for converting words
the chatbot and the features of the developed chatbot has not into vectors. The paper performs comparison of different
been discussed. Also, the paper proposes to develop an open machine learning classification algorithms in order to find the
ended chatbot system, but does not provide any details with class of the question asked. And, finally the query and answer
regards to the accuracy of the proposed system. mapping is done using cosine similarity. The paper claims to
produce an accuracy result of 87%.
Kumar Shivam, et. al. [19] propose to develop a chatbot
system for a college. The reason given for developing a The paper puts forth detailed elaboration of the
chatbot system of the college is that it is very hectic and technologies used for the development of the chatbot. But,
cumbersome for the enquiry staff to sit all day and answer the there are a few drawbacks of the proposed system. The paper
questions of the students. For the development of chatbot proposes to use very basic machine learning algorithms when
system the paper uses the facebook messenger application it is known that there are better deep learning algorithms as
programming interface which performs the natural language well. Also, for query and answer similarity cosine similarity
processing duties. The paper also proposes to use the method has been used other methods can be explored in order
pretrained module of wit.ai. to get better results.
The contribution of the paper is very limited in the field of
natural language processing as the paper makes use of III. ANALYSIS
standard API’s for the development of the chatbot system. The Table No. 1, 2 and 3 provides the analysis of the papers
inherent challenges in the natural language processing has not which have been reviewed in the previous section. This
been dealt with in the paper. analysis has been carried out with certain parameters like, the
Bingquan Liu, et. al. [20] are of the opinion that having the language processing technique used, chatbot system’s target
user’s personal information can greatly enhance the quality audience or domain of study, type of knowledge, machine
answers generated. In other words the paper proposes to take learning algorithm used, and processing of unstructured data
personal information of users into account while answering a by the system.
question so that the answers can be more personalised. By
giving personalised answers to the users it is expected that the
retention of the users will be good.
The paper proposes a useful idea of taking into
consideration the personal information while providing
answers to the asked questions. The paper also talks about
using Baidu Tieba Corpus (BTC) and Reddit Corpus (RC)
datasets. Apart from that the paper does not talk about any
technical details. The paper also does not give any details with
regards to the technologies used for the development of the
chatbot system nor does it talk about the type of knowledge,
whereas it is expected that the user’s personal information has
to be maintained in database.
On the other hand Takuma Okuda and Sanae Shoda [21]
propose to develop a chatbot framework for the financial
services industry. The paper provides the background for the
need for chatbot systems in the financial domain. The paper
also provides the different functions and features of the
chatbot framework and its uses in the financial domain. The
internal working of the chatbot with regards to the natural

978-1-7281-5197-7/20/$31.00 ©2020 IEEE 708

 2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS)

TABLE I. ANALYSIS OF CHATBOT SYSTEMS - 1

Paper Title Question Answer Automatic Question Smart Answering A Chatbot Using Automatic Dialogue
System for Online Generation from Chatbot based on LSTM-based Multi- Template Synthesis
Feedable New Born Children’s Stories for OCR and Layer Embedding for for Chatbot by Story
Chatbot [8] Companion Chatbot Overgenerating Elderly Care [11] Information
[9] Transformations and Extraction [12]
Ranking [10]
Language Processing NLTK framework Stanford tool Not mentioned Not mentioned CKIP toolkit and AIML
Technique used
Chatbot system’s Open-ended Children Open-ended Elders General
target audience or
Domain of study
Type of Knowledge Not mentioned Not mentioned Not mentioned Not mentioned Not mentioned
Machine Learning Not used Logistic Regression Not used LSTM Not used
algorithm used
Processing of No To a certain extent No No No
unstructured data by
the system

TABLE II. ANALYSIS OF CHATBOT SYSTEMS - 2

Paper Title Home Automation Real World Smart Chatbot for University A Novel Approach for Design and
using IoT and a Chatbot for Customer Related FAQs [15] Medical Assistance Implementation of
Chatbot using Natural Care using a Software Using Trained Interactive Product
Language Processing as a Service (SaaS) Chatbot [16] Manual System using
[13] Architecture [14] Chatbot and Sensed
Data [17]
Language Processing NLTK framework Not mentioned AIML Not mentioned Not mentioned
Technique used
Chatbot system’s Home Automation Customer service University / College / Medical Customer service
target audience or Education
Domain of study
Type of Knowledge Structured database Not mentioned Not mentioned Structured database Structured database
Machine Learning Not used Not mentioned which Not used Not mentioned which Not mentioned
algorithm used
Processing of No No No No No
unstructured data by
the system

TABLE III. ANALYSIS OF CHATBOT SYSTEMS - 3

Paper Title An Ergonomics Chatbot for College Content-Oriented AI-based Chatbot BANK CHAT BOT –
Evaluation to Chatbot Website [19] User Modeling for Service for Financial An Intelligent
Equipped with Personalized Response Industry [21] Assistant System
Knowledge-Rich Mind Ranking in Chatbots Using NLP and
[18] [20] Machine Learning
[22]
Language Processing AIML Facebook Messenger Not mentioned Not mentioned NLTK framework
Technique used Application
Programming Interface
Chatbot system’s General University / College / General Banking / Finance Banking / Finance
target audience or Education
Domain of study
Type of Knowledge Not mentioned Structured database Not mentioned Not mentioned Structured database
Machine Learning Not mentioned Not mentioned Deep Neural Network Not mentioned Random forest and
algorithm used SVM
Processing of No No No No No
unstructured data by
the system

978-1-7281-5197-7/20/$31.00 ©2020 IEEE 709

 2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS)
[7] Tatwadarshi P. Nagarhalli and Dr. N. K. Rana, “Review of Chatbot
IV. CONCLUSIONS Systems with Reference to Knowledge and Accuracy”, International
Journal of Advance and Innovative Research, Volume 7, Issue 1 (III):
Chatbot systems have become a very convenient way for January - March, 2020, pp. 1-7.
engaging with the customers. It has become a very convenient [8] Sameera A. Abdul-Kader and Dr. John Woods, “Question Answer
and interactive way through which, even complex, information System for Online Feedable New Born Chatbot”, IEEE Intelligent
can be passed along to the users in an effective manner. This is Systems Conference, London, UK, September 2017, pp. 863-869.
the reason because of which the chatbot systems have been [9] Che-Hao Lee, Tzu-Yu Chen, Liang-Pu Chen, Ping-Che Yang, Richard
adapted in so many different domains. Tzong-Han Tsai, “Automatic Question Generation from Children’s
Stories for Companion Chatbot”, IEEE International Conference on
From the review it can be seen that even though many Information Reuse and Integration for Data Science, 2018, pp. 491-494.
research papers have been written and the chatbot systems [10] Ly Pichponreay, Chi-Hwan Choi, Jin-Hyuk Kim, Kyung-Hee Lee, and
have been developed for many different domains, most of the Wan-Sup Cho, “Smart Answering Chatbot based on OCR and
Overgenerating Transformations and Ranking”, IEEE ICUFN, 2016, pp.
papers fail to mention the language processing technique used, 1002-1005.
the type of knowledge used, if any machine learning [11] Ming-Hsiang Su, Chung-Hsien Wu, Kun-Yi Huang, Qian-Bei Hong,
algorithms have been used or not, and whether unstructured Hsin-Min Wang, “A Chatbot Using LSTM-based Multi-Layer
data have been processed or not. The main focus of most of Embedding for Elderly Care”, IEEE International Conference on Orange
the papers remains towards the implementation and adaptation Technologies (ICOT), 2017, Singapore, pp. 70-74.
of the chatbot system than any research in the field of natural [12] Shih-Hung Wu, Liang-Pu Chen, Ping-Che Yang, Tsun Ku, “Automatic
language processing. Dialogue Template Synthesis for Chatbot by Story Information
Extraction”, IEEE International Conference on Information Reuse and
Many different machine learning or deep learning methods Integration for Data Science, 2018, pp. 485-490.
can be explored in order make the chatbot systems more [13] Cyril Joe Baby, Faizan Ayyub Khan, Swathi J. N., “Home Automation
accurate. As a matter of fact the basic natural language using IoT and a Chatbot using Natural Language Processing”, IEEE
International Conference on Innovations in Power and Advanced
processing tasks can be explored further in order to make the Computing Technologies [i-PACT2017], 2017, pp. 1-6.
data more suitable for further analysis. Also, the field of
[14] Godson Michael D’silva, Sanket Thakare, Sharddha More, and Jeril
processing of unstructured data, which is text data given as a Kuriakose, “Real World Smart Chatbot for Customer Care using a
knowledge is yet to be explored in an effective way. So, it can Software as a Service (SaaS) Architecture”, IEEE International
be concluded that there is ample scope for research in this area conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-
of natural language processing. SMAC 2017), pp. 658-664.
[15] Bhavika R. Ranoliya, Nidhi Raghuwanshi and Sanjay Singh, “Chatbot
for University Related FAQs”, IEEE International Conference on
Advances in Computing, Communications and Informatics (ICACCI),
2017, pp. 1525-1530.
References
[16] Divya Madhu, Neeraj Jain C. J, Elmy Sebastain, Shinoy Shaji, Anandhu
Ajayakumar, “A Novel Approach for Medical Assistance Using Trained
>@ Daniel Jurafsky and James H. Martin, “Speech and Chatbot”, IEEE International Conference on Inventive Communication
Language Processing: An Introduction to Natural Language and Computational Technologies (ICICCT 2017), 2017.
Processing, Computational Linguistics and Speech Recognition”, [17] Hanjong Choi and Takeshi Hamanaka, “Design and Implementation of
Prentice Hall Inc Publications, 2nd Edition, 1999. Interactive Product Manual System using Chatbot and Sensed Data”,
>@ Megha Desai, “Paving the way to a Chatbot Revolution”, published on IEEE 6th Global Conference on Consumer Electronics (GCCE 2017),
23rd Aug, 2019. Available at https://medium.com/searce/paving-the- 2017.
way-to-a-chatbot-revolution-c7ea21fa2f2f. Last accessed on 24th Jan, [18] Wei Liu, Jie Zhang, and Sheng Feng, “An Ergonomics Evaluation to
2020. Chatbot Equipped with Knowledge-Rich Mind”, IEEE 3rd International
>@ M. Milenkovic, “The Future is Now – 37 Fascinating Symposium on Computational and Business Intelligence, 2015, pp. 95-
Chatbot Statistics”, published on 30th Oct, 2019. 99.
Available at https://www.smallbizgenius.net/by-the-numbers/ [19] Kumar Shivam, Khan Saud, Manav Sharma, Saurav Vashishth and
th
chatbot-statistics/. Last accessed on 10 Dec, 2019. Sheetal Patil, “Chatbot for College Website”, International Journal of
>@ PTI, “Chatbots, Virtual Assistants Can Reduce a Manager's Workload Computing and Technology, Volume 5, Issue 6, June 2018.
by 70% by 2024: Gartner”, published 23rd Jan, 2020. Available [20] Bingquan Liu , Zhen Xu , Chengjie Sun, Baoxun Wang , Xiaolong
at https://www.news18.com/news/tech/chatbots-virtual-assistants-can- Wang, Derek F. Wong and Min Zhang, “Content-Oriented User
reduce-a-managers-workload-by-70-by-2024-gartner-2470345.html. Modeling for Personalized Response Ranking in Chatbots”, IEEE/ACM
Last accessed on 25th Jan, 2020. Transactions on Audio, Speech, and Language Processing, Vol. 26, No.
>@ Editors at BRN.AI, “Chatbot Report 2019: Global Trends and Analysis”, 1, January 2018, pp. 122-133.
published on 19th Apr, 2019. Available at [21] Takuma Okuda and Sanae Shoda, “AI-based Chatbot Service for
https://chatbotsmagazine.com/chatbot-report-2019-global-trends-and- Financial Industry”, Fujitsu Science and Technology Journal, Vol. 54,
analysis-a487afec05b. Last accessed on 10th Dec, 2019. No. 2, April 2018, pp.4-8.
>@ IANS, “Chatbots at work: Consumers prefer interacting with voice [22] Chaitrali S. Kulkarni, Amruta U. Bhavsar, Savita R. Pingale, Prof. Satish
assistant, not humans”, published 5th Sept, 2019. Available S. Kumbhar, “BANK CHAT BOT – An Intelligent Assistant System
at https://economictimes.indiatimes.com/magazines/panache/chatbots- Using NLP and Machine Learning”, International Research Journal of
at-work-consumers-prefer-interacting-with-voice-assistant-not- Engineering and Technology (IRJET), Volume: 04 Issue: 05, May -
humans/articleshow/70993297.cms?from=mdr. Last accessed on 2017, pp. 2374-2377.
25th 2020.
Jan,

978-1-7281-5197-7/20/$31.00 ©2020 IEEE 710