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

net/publication/356487897

Modern Development Trends of Chatbots Using Artificial Intelligence (AI)

Conference Paper · October 2021

DOI: 10.1109/ITMS52826.2021.9615258

CITATIONS READS

22 2,629

6 authors, including:

Julija Skrebeca Andrejs Romanovs

Riga Technical University Riga Technical University
2 PUBLICATIONS 24 CITATIONS 85 PUBLICATIONS 772 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Andrejs Romanovs on 09 December 2021.

The user has requested enhancement of the downloaded file.

 Modern Development Trends of Chatbots Using
2021 62nd International Scientific Conference on Information Technology and Management Science of Riga Technical University (ITMS) | 978-1-6654-0615-4/21/$31.00 ©2021 IEEE | DOI: 10.1109/ITMS52826.2021.9615258

Artificial Intelligence (AI)

Julija Skrebeca Paula Kalniete Janis Goldbergs
Institute of Information Technology Institute of Information Technology Institute of Information Technology
Riga Technical University Riga Technical University Riga Technical University
Riga, Latvia Riga, Latvia Riga, Latvia
julija.skrebeca@gmail.com paula.kalniete@inbox.lv j.goldbergam@gmail.com

Liene Pitkevica Darja Tihomirova Andrejs Romanovs

Institute of Information Technology Institute of Information Technology Dept. of Modelling and Simulation
Riga Technical University Riga Technical University Riga Technical University
Riga, Latvia Riga, Latvia Riga, Latvia
liene.pitkevica@edu.rtu.lv darjatihomirova3@gmail.com andrejs.romanovs@rtu.lv

Abstract—It is said 80% of business owners will use chatbots The term chatbot – chat(-ter) bot was invented by Michael
in the future. Chatbots become more and more popular in terms Loren Mauldin. According to Shawar and Atwell’s simplified
of business, that is why it is necessary for businesses implement definition, chatbots are artificial intelligence-assisted chat
innovative approaches to provide customer service twenty-four applications whose functions range is from answering simple
hours. Such business especially is needed in terms of the questions to participating in complex conversations [10].
challenging Covid-19 times. Artificial Intelligence-powered
chatbots can work as intelligent teaching systems, for providing A chatbot is a software application that helps to carry on
a personalized way of learning for students. Chatbot reviews a conversation using text-based or auditory methods [5].
student's responses and his learning progress. One of the most Programs in Chatbots are developed to mimic human
convenient features of chatbots is the opportunity to send lecture conversations. Chatbots are used for variety of purposes,
materials in the form of messages to students as if it is just a chat including customer service, routing request, and information
with a friend. Apart from personalized chatbot usage in the retrieval. By using Natural language processing, some
studying process, it can be used to streamline business processes, chatbots can be used for word classification.
e.g., such as sales. Nowadays chatbots are able to help customers
to search products they need, place orders to the cart and pay The simplest botts processed basic user messages and
for it, and track delivery processes of orders. Chatbot requests. In a posted conversation with the user, with an
intelligence is being developed every day and every year now, so algorithm, it responds with pre-programmed answers.
very soon chatbots will be able to perform even more difficult Chatbots are most frequently used by dialog systems, such as
tasks to make the life of the user easier. In this paper, authors customer support.
will deliver theoretical materials and historical background of
chatbots, describe classification and techniques of chatbots, then Artificial Intelligence (AI) simulates human intelligence
describe the modern development trends of chatbots using in the form of various devices. Such AI devices are developed
artificial intelligence (AI), and finally discuss the role of chatbots to behave as humans do and imitate their actions. Artificial
in education and e-commerce. Intelligence stands for machines that exhibit natural
Keywords— artificial intelligence (AI), chatbots, covid-19, e- characteristics similar to the human mind, such as problem
commerce, education, trends solving and understanding.
I. INTRODUCTION Natural Language Processing (NLP) is the program
interface that allows computers and humans
A chatbot can be represented as a computer, program, intercommunicate with each other. NLP makes it possible for
algorithm, or artificial intelligence which the main goal is a computer to analyze vast amounts of language data from
communication with a person, or another interlocutor. different sources. NLP technique developers can improve the
Chatbots might be developed to communicate by replying to chatbot knowledge to process various tasks such as text
simple keywords or to interact with a user by supporting analysis, stemming, text summarization, automatic
specific topics. They can be used for a customer service, summarization, topic extraction, text mining, speech
marketing, advertising, entertainment industry, data collection recognition, translation, segmentation, and automatic question
etc. For example, a chatbot can communicate with customers answering.
in online shops and help them to find the product they need. NLP combines two techniques – natural language
In this paper, authors have done research in chatbot history generation (NLG) and natural language understanding (NLU).
to find out how the first chatbots were made, which methods NLP is one of the forms of the Artificial Intelligence that
were used, and how much chatbots have been enhanced in the enables chatbots to process users’ text or audio messages and
last few decades. Also, the authors have referred to the most give a proper answer. NLP has the following layers:
common methods and techniques that are used to create a • Application.
chatbot. In this paper, there is mentioned classification of
chatbots and offered both classification and methods to actual • Data Storage.
chatbot trends in the Covid-19 era.
• NLP Engine.

978-1-6654-0615-4/21/$31.00 ©2021 IEEE

 • Platform for Data Lake. example was written by Racter: “A crow is a bird, an eagle is
a bird, a dove is a bird. They all fly in the night and in the day.
NLP contains two important techniques to work properly: They fly when the sky is red and when the heaven is blue.
• NLU – Natural Language Understanding maps user They fly through the atmosphere. We cannot fly. We are not
given input to its useful representations. It analyzes like a crow or an eagle or a dove. We are not birds. But we can
unusual phenomena of language. dream about them. You can.” [1].

• NLG – Natural Language Generation is used for text The same method was used with a chatbot called
and sentence planning, mining, and text realization. Jabberwacky designed to mimic a natural human conversation
using a voice-activated system.
NLU is the process of generating meaningful responses in
the form of natural language, and includes: Another milestone in the development of chatbots was
called Dr. Sbaitso chatbot. It was able to synthesize speech –
• Text planning retrieves important data from the in some ways it became even more human-like than its
database. predecessors, even if it could not converse in a more
complicated and complex way. Such problem of chatbot
• Sentence planning selects the required words, produces
communication insufficient complexity still exists today.
relevant phrases using these words and creates
meaningful sentences. Kuki - one more chatbot that needs to be represented
(formerly known as Mitsuku) that claims to be a young girl of
• Text implementation is the flow mapping of the 18 years from England [9]. Kuki is developed using AI Mark-
sentence according to the sentence structure. Up Language and both - Retrieval and Generative-Based
It is much harder to realize Natural Language approaches which allows her to imitate a human conversation,
Understanding process than execute the Natural Language avoiding answering the same question over and over again.
Generation process. The most popular chatbots at the time are chatbots that
Natural Language Toolkit (NLTK) was programmed for are used for e-commerce and e-education solutions. There
the creation and application of symbolic and statistical NLP in have always been difficulties in communication between
Python. It is a pack of libraries that include parsing and customer and seller or student and teacher as sometimes there
classification of text, text processing for tokenization, might be situations where employees' workload is huge and
stemming of words, and reasoning about semantics. answering all these questions can cause a big delay although
most probably the question can be answered quickly, and the
Keras is an API library for neural networks in Python. It information might be found easily. This is where chatbots can
combines well with R, PlaidML, TensorFlow, and the be helpful to reduce communication problems in terms of
Microsoft Cognitive Toolkit. response time. An e-commerce chatbot can assist to find the
Tkinter is an interface for developing graphical user product that a customer is searching for or help to fill a parcel
interface (GUI) applications. returning template. For students, chatbots can answer
questions that are related to due dates or requirements in a
Tensorflow is a software library framework that focuses specific subject. Chatbots are very common nowadays and
mainly on machine learning. It uses data flow graphs and they are going to improve and become more helpful and
differentiates programming across different numbers of tasks necessary in the future.
to build models. It is used to create large-scale development
applications including neural networks. It is mainly used for Currently, there are plenty of well-known chatbots that are
classification, understanding, prediction, and creation. used in daily life. For example, Siri – a virtual assistant of iOS
interface that uses natural language to answer user inquiries
Nowadays there are many chatbots, but how did they start? and perform web service requests. Google Now is another
What was the first chatbot? mobile application for Android and iOS devices that provides
Eliza is one of the most famous and at the same time oldest Google-developed predictive maps with information and daily
chatbots invented by the Artificial Intelligence Laboratory in updates to automatically answer users’ questions. Cortana -
MIT, dating back to 1964. Eliza inspired many developers in the Windows voice platform with information that helps
the field to develop their own chatbots. In the early scenario software and hardware developers. Chatbots for Messenger -
called DOCTOR, the chatbot Eliza has the role of a Rogerian a program that uses Artificial Intelligence (AI) to realize an
psychotherapist, asking open-ended questions, which she also interaction with customers. Facebook launched the platform
answers, drawing attention from herself to the user. to allow developers to produce chatbots that are able
communicate with Facebook users through the Messenger
PARRY is another well-known chatbot developed in chat interface. Messenger chatbots understand users’
1972. This program, which diverts attention from itself, uses questions and formulate responses in a very human way [3].
an opposite strategy than Eliza. It does not behave like a doctor
but like a paranoid schizophrenic patient. It tries to provoke Nowadays, functional chatbots are no longer a rarity. They
controversy and makes the interlocutor give more detailed are being developed more and more advanced. One more
answers [1]. interesting project is Sophia, the chatbot that hides behind the
appearance of a humanoid female robot with highly advanced
Racter is another knowable chatbot (short for raconteur – facial expressions [1].
a storyteller). This Chatbot generates prose in the English
language. Prose can be a short text, not written in the verse, in Interesting research was done by a group of students from
which the authors present fragments of their existential the Philippines - chatbots also can help to deal with emotional
experience. This chatbot uses logical sentences based on its traumas during rehabilitation or as a prevention of an
knowledge and makes conclusions one by one. A good emotional breakdown by simulating empathy [6]. Such
chatbots have two types of perceived input data - typed text
and user’s facial expression transmitted via video. For now,
such chatbots powered by Deep Learning can recognize
neutral, happy, and sad users’ facial expressions with 16.7%,
66.7%, and 56.7% accuracy respectively. The combination of
textual and video data input provides a chatbot with sufficient
information to create a proper answer in a certain
conversation. So, it is even possible that in the future chatbots
will be able to assist psychologists or even psychiatrists during
patient treatment.

II. CHATBOT CLASSIFICATION AND TECHNIQUES

Chatbots are classified (Fig. 1) by their interaction, Fig. 1. Classification of chatbot development approaches
information gathering, and usage goal types, e.g., Text-Based
and Voice-Based for interaction mode, Open Domain and There are Rule-Based and Self-Learning (includes
Closed Domain for knowledge domain, Task-Oriented and Generative-Based, Retrieval-Based Approaches) chatbots [7],
Non-Task-Oriented for goals, and Rule-Based, Retrieval- [9]. Any of the two Self-Learning chatbot approaches can
Based, Generative-Based for design approach [7], [9]. Some have different techniques such as:
of them are shortly described further. • Parsing method chooses exact words from the input
Task-Oriented chatbots (e.g., Alexa, Cortana, Siri) goal text, makes them less sophisticated, and uses them for
is to help a user to fulfill his task. Task-Oriented chatbots work output.
in restricted domains and are made to help with a hotel or • Pattern Matching method takes as a pattern user’s
flight booking, create a schedule, or find specific information, input and gives him the most suitable response stored
etc. in the template.
Non-Task-Oriented chatbots can handle an extended • Artificial Intelligence Mark-up Language is made of
conversation with the user, which helps to create a feeling of data objects (AIML elements) that consist of topics and
interaction with a human. These responses of chatbots can be categories. Categories have patterns (matches the user
divided into two approaches – Generative-Based (generates input) and templates (creates an output) which provide
more proper responses during the conversation) and Retrieval- a flexible conversation.
Based (learns to select more informative responses from a
repository of a current conversation). None of them can • Markov Chain Model uses probabilities to know
answer if there is no pattern for that question, because they do what kind of input data will be given to a chatbot next.
not create responses, but only take predefined ones. Mostly used for imitating simple human conversation.
Cannot be used for an extended conversation in a
Domain-Specific chatbots, which include Open-Domain specific area.
(used for a non-specific conversation) and Closed-Domain
(conversation has a specific goal) Approaches, are used in • Artificial Neural Networks Models were made to
specific areas (e.g., education and health care) which helps to develop smarter chatbots and can use both -
raise their efficiency and quality of given answers. Generative-Based and Retrieval-Based approaches.
Such chatbots learn how to interact from conversations
Text-Based and Voice-Based chatbots differ by their with a human. Deep Learning allows imitating a
interaction type with users. Speech-to-text is one of the most human brain's work to process given data and produce
crucial functions in any competitive chatbot and its new patterns.
improvement. The basic idea of speech-to-text consists of
criteria like vocabulary size – the number of words in Even though none of all chatbots passed the Turing Test
vocabulary which are millions from different languages. since the first one (ELIZA) had been invented and chatbots
Another criterion is speaker independence or in other words based on Deep Learning are not an exception, nowadays the
chatbot’s ability to recognize speakers. Co-articulation is an most promising technique for chatbot development is Deep
important part of speech-to-text because a chatbot must have Learning. It is believed that Deep Learning combined with a
the ability to process a continuous stream of words that Retrieval-Based Approach can improve the chatbot
requires segmentation and tokenization of the speaker’s input. intelligence and give results that will show a significant
The chatbot must handle noise to filter out background noises difference compared to previous results.
and must be able to understand input when a person is talking
at different distances from the microphone [4]. III. CONTINUOUS IMPROVEMENT STUDY OF CHATBOT
TECHNOLOGIES USING A HUMAN FACTORS METHODOLOGY

A usability test was conducted to compare the usability of

three chatbot platforms – Watson by IBM, Pandorabot by
Pandora, and Verbot [2].
The research study was done by using these methods:
• Gathering demographic information of participants.
• Pre-test questionnaire.
 • Video and/or audio record chatbot sessions. • Chatbots are able to help businesses save up to 30% on
customer support costs.
• Post-test questionnaire.
• The market value of chatbots was $703 million in
For the purpose of the research study, the feedback was 2016.
obtained from ten participants and then assessed using a
System Usability Scale (SUS). Results indicated that Watson • More than 50% of customers expect businesses to be
by IBM was perceived as the most user-friendly platform open and answer their questions 24/7.
overall. Watson received an average SUS score of 81.875 out
of 100, Pandorabot scored 88.75 out of 100. Verbot did not • Chatbots are popular among both – Millennials and
receive a SUS score as far as none of the ten participants chose Baby Boomers. There are more than 300,000 active
this platform. While Pandorabot scored higher at SUS, 80% chatbots on Facebook.
of the participants preferred Watson platform. The statistical Exist four main issues that contribute to skepticism about
discrepancy between these participants’ responses was the chatbot usefulness. Quite simply, chatbots cannot execute
attributed to the fact that only a small group of participants technical commands that the user types. Unfortunately, for
chose Pandorabot. Research results showed that Watson by now chatbots are unable to properly process a customer’s
IBM is the chatbot that best matches with human factors intent, resulting in misinterpreted queries and responses.
analysis. Watson chatbot had perceived intelligence, a Chatbots do not have enough of conversational intelligence,
simplified atmosphere, and was chosen by 80% of meaning they often do not process the implied details of a
participants. dialog, resulting in an inadequate conversation. As well,
chatbots are unable to understand different accents and
IV. CHATBOT STATS AND TRENDS SHAPING BUSINESSES IN cultural meanings to provide a correct response. AI chatbot
2021 trends are poised for profound change that will impact several
key business processes. These processes include:
In many service industries chatbots are used to answer
customers’ questions and help them navigate a company’s • Automation of business processes.
website. Well-developed and sophisticated chatbots rely on
• Prediction of costumer behavior.
machine learning, which is a computer program that
continually improves through its usage, and Natural Language • Recommendation of products and services.
Processing (NLP), which helps solve the problems of
mimicking human-generated text and speech. • Streamlining customer support experiences [16].
Chatbots that feel increasingly humanlike are already The increasing expectations of twenty-four hours chatbot
being widely deployed. It all happens thanks to Natural availability means that chatbots must be able to analyze user’s
Language Processing (NLP). NLP allows chatbots to interact speech, text, and even facial expressions. Eventually, chatbots
with users using in complete sentences that have a natural will be used in a variety of consumer applications and internal
conversational flow. Dialects, small sounds, deliberate pauses, business functions [8], [15].
or even misspelling can help customers feel more
V. CHATBOT ROLE IN EDUCATION
comfortable. Like that conversation seems to be more realistic
and lifelike. This, in turn, makes customer service easier for
everyone involved and helps companies further improve The use of chatbots in education would make life easier
customer experience and loyalty [14]. during Covid-19. When all training in the educational
establishment takes place in remote mode. The use of
It is important to avoid any misunderstanding when a chatbots could send statements about exams and systems
chatbot must give correct and important information. The could help with the understanding of the teaching substance.
main chatbot challenges are: The Institute of Technology in Georgia, USA received
• Misunderstanding requests. Chatbots often around 10,000 questions about the conduct of training. The
misunderstand users’ requests because they are unable number of questions was difficult to answer in time. So,
to understand the will of the consumer. Ashok Goel invented the chatbot Jill. Chatbot answered many
different questions: about the format of the exam, possible
• Inaccurate execution of commands. Chatbots cannot topics, and deadlines. Teachers answered difficult questions,
respond to any technical commands issued by
but an algorithm was invented for simple questions. The
customers.
chatbot was trained on 40,000 specific issues raised over the
• Difficulty understanding accents. Chatbots are not yet past few years. When the answers generated 97% accuracy,
able to understand accents or dialects to identify the Chatbot went online [10].
correct intent of the user [13].
VI. CHATBOTS ROLE IN E-COMMERCE AND EVALUATION
AI technology is evolving every day, which suggests that METHODS
chatbots are also subjects that are in a constant state of change.
Nowadays, there are some chatbot trends in business and other E-Commerce has grown in Latvia during Covid-19. This
areas. The statistics show that the healthcare, finance, travel,
has significant implications for society and business
education, and real estate industries are benefiting the most
from chatbots: worldwide. Consumers are attracted to the fact that there is a
wider choice of goods at much lower prices than customers
• By 2021 about 80% of businesses are expected to would pay in a local shop, as well they are attracted by the
integrate some form of chatbot system. idea that they can shop anywhere all around the world, taking
advantage of exchange rates and economic differences. But
there is also a significant disadvantage, e.g., many customers, Chatbots for educational processes (Fig. 2) should have
mostly people from older generations, do not trust online knowledge in a specific area be it math, physics, and
shopping on the Internet, because they cannot check the chemistry, or literature, music, and geography.
quality of the product and ask about the warranty [11]. The
chatbot would help preserve the cash of various large e-shops,
by starting to use advisers to answer client questions. And
therefore, it is crucial to be able to determine how good your
chatbot is. There is a wide variety of chatbot evaluation
methods that determine the performance of commercially
available chatbots. This paper describes two - PARADISE
and Kuligowska. PARADISE estimates subjective factors
such as ease of usage, clarity, naturalness, friendliness,
robustness reading misunderstanding, and willingness to use
the system again. This framework maximizes task
effectiveness and minimizes dialogue cost by determining
overall performance:

𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 = ∝ ∗ 𝑁𝑁(𝑘𝑘) − ∑𝑛𝑛𝑖𝑖=1 𝑊𝑊𝑖𝑖 ∗ 𝑁𝑁(𝐶𝐶𝑖𝑖 )

where α is the weight on k, each cost function is weighted by

Wi and N is a Z score normalization function. Kuligowska
method measures the chatbot in 8 categories – visual look,
implementation way of chatbots, either it is a built-in Fig. 2. Educational Chatbot structure
window, pull out tab or both, giving the highest score for both
GUI. This framework quantifies the bot’s ability to produce Closed-Domain class chatbots, which are used for
speech, either it has a unique custom voice, has text-to-speech conversations in specific areas, fit such descriptions
modules without custom voice, or can communicate only perfectly. The Retrieval-Based approach of Self-Learning
with text, giving the highest score for bots with custom voice chatbots will help an educational chatbot to specialize better
modules. The fourth category is both abilities to answer and generate much more proper responses than Generative-
domain-based questions like “What products are you Based chatbots that tend to create wrong or illogical
selling?”. Are chatbots able to lead a coherent dialogue and responses due to its complexity. Educational chatbots must
understand the dialogue’s context [4]? give adequate and accurate information in the context of the
It is important to not just build and launch a chatbot but educational process, even if communication will be based on
to be sure that the chatbot is engaging and helpful, otherwise, predefined and repeated phrases. As a technique, the AI
buyers would not interact with the chatbot. Statistics say that Mark-Up Language - has its own data architecture that
about 40% of users will stop interacting with a bot after the consists of objects, topics, categories, response patterns, and
first message, and 25% of users after the second. To avoid templates which is very similar to the educational process,
these poor statistics, it is necessary to measure chatbot where students have different subjects and topics. Each
activity to make sure that the bot is benefiting buyers and category has a special rule that matches the user’s input and
driving sales. This analytic data includes basic metrics that allows a proper answer.
indicate the bot’s helpfulness, like retention rates, advanced Even though E-commerce chatbots have a little bit
metrics, and engagement rates to measure sales indication. To different goals and direction of usage, they can have almost
be able to collect and analyze sales data, there are several the same structure as education chatbots have (Fig. 3). It
options on how to achieve it, like using analytics tools in should be a Closed-Domain class chatbot that will be able to
chatbot building platforms or integrating chatbot to an consult clients in any E-commerce area starting with clothes
external analytics platform [12]. and electronic equipment and ending with trading and
insurance. Closed-Domain class allows to have knowledge in
VII. CHATBOT DEVELOPMENT TECHNIQUE a specific area which will enlarge chatbot usage benefits and
increase clients’ satisfaction. E-commerce chatbots need to
Since 1964, when the first chatbot Eliza was invented, be Self-Learning with a Generative-Based approach to adapt
chatbots, their classifications, and architecture have been to purchase tendencies and know the user's preferences to
developing dramatically. During Covid-19 pandemic times offer better products for a certain client. Even though
chatbots became even more popular than before due to the Generative-Based chatbots sometimes tend to
massive digitalization of many industries such as E- “misunderstand” their interlocutor and can make a mistake
commerce, finance, health care, education, traveling, etc. while responding to the client, it is very necessary for E-
The research showed the importance of chatbots in commerce chatbot to give different answers as far as they are
everyday life, so it is very necessary to use their capabilities not predefined to give client a feeling of communication with
to make our lives better and easier during this tough pandemic a real consultant. Due to a wide selection of goods, the
time. Chatbots for E-commerce and education will be very chatbot must “see” differences between two similar products
useful for businesses and students all around the world, but to offer a better one that matches the client's preferences the
such chatbots must have special approaches as far as they most. Deep Learning can help to provide such a personalized
have different goals of usage. attitude towards a customer due to its ability to create its own
“opinion”, which can help to provide a client with a product research on chatbot trends connected to the health care and
or service that client needs. chatbot users’ emotional state.
As a result of the survey were defined work principles for
educational and E-commerce chatbots. Both are Self-
Learning chatbots, but use different approaches, and
developed using AILM (for educational chatbot) and Deep
Learning (for E-commerce chatbot) techniques which fit the
goal of each chatbot.
REFERENCES

[1] T. Zemčík, “A Brief History of Chatbots,” VŠB-Technical University

of Ostrava, Czech Republic, 2019.
[2] K. Coperich, E. Cudney, H. Nembhard, “Continuous Improvement
Study of Chatbot Technologies using a Human Factors Methodology,”
Industrial and Systems Engineering Conference, 2017.
[3] A. A. Qaffas, “Improvement of Chatbots Semantics Using Wit.ai and
Word Sequence Kernel: Education Chatbot as a Case Study,”
International Journal of Modern Education and Computer Science
(IJMECS), MECS, 2019.
[4] J. Cahn, “CHATBOT: Architecture, Design, & Development,” Senior
Thesis, School of Engineering and Applied Science, Pennsylvania,
USA, 2017.
[5] M. Dharani, J.V.S.L. Jyostna, E. Sucharitha, R. Likitha, S. Manne,
“Interactive Transport Enquiry with AI Chatbot,” 4th International
Fig. 3. E-commerce Chatbot structure Conference on Intelligent Computing and Control Systems, IEEE,
2020.
As an example of a well-developed chatbot was used [6] K. A. L. R. Carranza, J. Manalili, N. T. Bugtai, R. G. Baldovino,
Kuki (former Mitsuki) which combines Retrieval and “Expression Tracking with OpenCV Deep Learning for a Development
Generative-Based approaches and uses AI Mark-Up of Emotionally Aware Chatbots,” 7th International Conference on
Robot Intelligence Technology and Applications, IEEE, 2019.
Language. But the usage of Generative-Based approach in
[7] S. Hussain, O. A. Sianaki, N. Ababneh, “A Survey on Conversational
Educational chatbot will not be that suitable due to this Agents/Chatbots Classification and Design Techniques,” 33rd
approach unstable work. E-commerce chatbot will be less International Conference On Advanced Information Networking And
interesting and useful for clients as well, if it will be Applications, Springer, 2019.
developed using the AI Mark-Up Language method due to its [8] Team Linchpin, “25 Chatbot Stats and Trends Shaping Businesses in
inflexibility. 2021,” LinchpinSeo, 2021. [Online]. Available:
https://linchpinseo.com/chatbot-statistics-
To sum up, chatbot classes can differ due to goals which trends/#:~:text=Chatbot%20trends%20in%202021,-
chatbots must reach, but nowadays the most popular chatbots Experts%20believe%20that&text=47%25%20of%20organizations%2
are the Self-Learning ones, which can offer dynamic and 0are%20expected,services%2C%20and%20improve%20customer%2
0service. [Accessed 19 10 2021].
human-like communication. Among techniques the most
[9] S. Singh, H. K. Thakur, “Survey of Various AI Chatbots Based on
progressive and promising are connected to AI, e.g., Deep Technology Used,” 8th International Conference on Reliability,
Learning, which is used in many different areas, has all Infocom Technologies and Optimization (ICRITO), IEEE, 2020.
chances to develop an improved chatbot performance so they [10] G. Molnar, Z. Szuts, “The Role of Chatbots in Formal Education,” 6th
can seem to be humans during the conversation. International Symposium on Intelligent Systems and Informatics,
IEEE, 2018.
VIII. CONCLUSION [11] M. M. Khan, “Development of An e-commerce Sales Chatbot,” 17th
International Conference on Smart Communities: Improving Quality of
E-commerce and education chatbots have different usage Life Using ICT, IoT and AI (HONET), IEEE, 2020.
goals. In the educational process chatbots help with [12] M. Ugino, “Everything You Need to Know About Chatbots in
Ecommerce,” Sellbrite, 07 02 2018. [Online]. Available:
organizational issues, e.g., they provide students with https://www.sellbrite.com/blog/ecommerce-chatbots/. [Accessed 19 10
information about the format and date of exams, possible 2021].
milestones and their deadlines. In E-commerce chatbots help [13] S. Patel, “Top 12 Chatbots Trends and Statistics to Follow in 2021,”
customers by answering their questions, e.g., inquire about REVE Chat, 13 08 2021. [Online]. Available:
https://www.revechat.com/blog/chatbots-trends-stats/. [Accessed 19
fabrics, ask about delivery duration etc. Even though chatbots 10 2021].
can have different goals of usage they can be similar in [14] R. Karnaukh, “6 Chatbot Trends that Are Bringing the Future Closer,”
structure. Chatbots can belong to a closed-domain class ONIX. [Online]. Available: https://onix-systems.com/blog/6-chatbot-
which can provide a person with information in any area. For trends-that-are-bringing-the-future-closer. [Accessed 19 10 2021].
better performance chatbots are better to be self-learning with [15] I. Sitova, J. Pecerska, “Process Data Analysis Using Visual Analytics
a generative approach to adapt to human needs. So, the thing and Process Mining Techniques,” 61st International Scientific
Conference on Information Technology and Management Science of
that differs the most often are approaches of chatbots. Due to Riga Technical University (ITMS), IEEE, 2020.
the big number of existing approaches, chatbot developers [16] Y. E. Midilli, S. Parsutins, “Optimization of Deep Learning
can choose any that will fit the goal of communication with a Hyperparameters with Experimental Design in Exchange Rate
chatbot. It is worth saying that the most promising approach Prediction,” 61st International Scientific Conference on Information
for chatbot improvement is Deep Learning that can make the Technology and Management Science of Riga Technical University
(ITMS), IEEE, 2020.
far future closer. In the future it is planned to continue the

View publication stats