=Paper=
{{Paper
|id=Vol-2407/paper-02-164
|storemode=property
|title=
A chatbot as an environment for carrying out the group decision making process
|pdfUrl=https://ceur-ws.org/Vol-2407/paper-02-164.pdf
|volume=Vol-2407
|authors=Olga Chukhno,Nadezhda Chukhno,Konstantin Samouylov,Sergey Shorgin
|dblpUrl=https://dblp.org/rec/conf/ittmm/ChukhnoCSS19
}}
==
A chatbot as an environment for carrying out the group decision making process
==
https://ceur-ws.org/Vol-2407/paper-02-164.pdf
15
UDC 519.816
A chatbot as an environment for carrying out the group
decision making process
Olga Chukhno* , Nadezhda Chukhno* ,
Konstantin Samouylov*† , Sergey Shorgin†
*
Peoples’ Friendship University of Russia (RUDN University),
6 Miklukho-Maklaya St, Moscow, 117198, Russian Federation
†
Federal Research Center “Computer Science and Control”
of the Russian Academy of Sciences (FRC CSC RAS),
44-2 Vavilov St, Moscow, 119333, Russian Federation
Email: olgachukhno95@gmail.com, nvchukhno@gmail.com, samuylov_ke@rudn.university, ssorgin@ipiran.ru
Since the early 1960s, people have been interested in creating a robot that can communicate
with humans (for example, ELIZA, ALICE and SmarterChild). Therefore, it’s not surprising
that chatbots are an incredibly sought-after topic now. This trend is due to the growing
popularity of messaging platforms, as well as the development of artificial intelligence and
machine learning. Firstly, the way of communicating is radically changed. We moved from
phone calls to text messages, and then to messaging applications. Secondly, the current state
of the field of artificial intelligence, natural language processing (NLP) and voice recognition
allow to robots to understand user’s requests better and respond to them accordingly. In fact,
chatbots are becoming new interfaces in messaging applications, replacing many other mobile
apps. Some experts argue virtual assistant of this type will be able to replace 99% of software
in the near future. Thus, instead of developing a new application, it becomes possible to
create a service that works in an already installed platform on the user’s device. The goal
of the work is to present a solution in which users will be able to participate in the GDM
process using natural language at any time and any place. Moreover, the chatbot will make
easier to provide experts preferences using an apprehensible interface. The paper proposes
a virtual interactive system developed for implementing the method of ranking alternatives
in the group decision making process. The designed digital assistant works on the base of a
cross-platform messenger Telegram.
Key words and phrases: group decision making process, chatbot, conversational sys-
tems, machine learning, decision support system, mobile application.
Copyright © 2019 for the individual papers by the papers’ authors. Copying permitted for private and
academic purposes. This volume is published and copyrighted by its editors.
In: K. E. Samouylov, L. A. Sevastianov, D. S. Kulyabov (eds.): Selected Papers of the IX Conference
“Information and Telecommunication Technologies and Mathematical Modeling of High-Tech Systems”,
Moscow, Russia, 19-Apr-2019, published at http://ceur-ws.org
16 ITTMM—2019
1. Introduction
The group decision making process (GDM) in fact is choosing the best options from
an acceptable set of alternatives and is presented in almost every task of daily life. The
use of new technologies expands decision making capabilities and allows these processes
to be conducted in situations that were previously impossible [1]. For instance, now
experts from different countries of the world may be involved in one process of decision
making. However, even with the development of mobile technologies, there is still an
important need for new interaction tools between the experts and providing information
to the system in cases where they can not meet in one place at the same time [2–5].
The implementation of mobile technologies in the GDM processes is based on the
assumption that it will lead to an optimization of the group decision making process,
namely, experts will be more focused on the problem and will be able to spend less time
providing their assessments to the system [6]. These tasks differ from optimization
problems traditionally solved the authors’ research team, for example, optimizing the
functioning of wireless telecommunication systems depending on interference [7, 8], or
developing hysteresis network overload control mechanisms [9, 10].
The recent boom in artificial intelligence has increased people’s attention to con-
versational entity, commonly referred to as chatbots [11–13]. Chatbot is a virtual
assistant that is able to interact with the user using natural language. As a rule, it can
ask questions to the user and answer them, suggest a topic to be discussed, etc. In
the modern world, chatbots have been implemented for education, information search,
customer service, communication, site navigation, analytic, design, developer tools,
games, health, personal, productivity, HR, marketing, news, shopping, social, sports,
travel, there are even bots that provide entertainment services [14–16]. However, we
assume that in the field of group decision making, our work will be the first attempt to
create a conversational program that has the ability to communicate with experts 24/7
hours (Fig. 1).
Figure 1. Chatbot’s domains
The GDM method that can work with a large amount of information have been
presented in [6]. The method of alternatives ranking was developed to carry out the
process in social networks, when the number of experts involved in the process was large.
Chukhno O. et al. 17
However, the procedure for providing assessments by experts was not automated and
this step of the process took a lot of time over the experts.
The purpose of the paper is to develop a prototype of a mobile virtual assistant that
automatically processes the answers of experts. Smartbot will provide a new approach to
working with a dynamic group of experts (during the process the number of participants
may change), making decisions about alternatives at any place and at any time. In fact,
at each stage of the group decision making process, users:
– will be informed about a set of alternatives that need to be compared in pairs;
– will be able to select independently a scale for evaluating alternatives, guided by
the degree of confidence regarding the level of knowledge about the problem/topic
under consideration;
– will be able to participate in the process anytime and anywhere by connecting to
the Internet.
Accordingly, developing the tasks posed in [6], the authors of this article create
an environment for more extensive numerical experiments of the method, which will
improve indicators such as the degree of participation of experts in the GDM process
and guaranteed to increase the reliability of the results.
The rest of the paper is organized as follows. Section 2 provides an overview that is
necessary for understanding the group decision making process, as well as explaining to
the reader what chatbots are, why they are needed, and why these interactive agents are
so popular nowadays. The third section describes the system architecture. The chatbot
implementation for the GDM process is presented in the fourth section. The fifth part of
the paper reveals the main advantages of introducing an artificial conversational entity
into the area of group decision making. In conclusion, results are summarized and the
tasks for further research are set.
2. Preliminaries
2.1. Ranking Method of Alternatives for the Group Decision Making
Process
The group decision making process can be defined as follows [2, 4]: the set of experts
𝐸 = {𝑒1 , ..., 𝑒𝐾 } should present their preferences regarding to the set of alternatives
𝑋 = {𝑥1 , ..., 𝑥𝑀 }, where |𝐸| = 𝐾 < ∞ and |𝑋| = 𝑀 < ∞.
The goal of the group decision making process is to order different alternatives
from a set of alternatives from the best to the worst with the help of the association
of some preference degrees, taking into account the opinions expressed by group of
decision-makers.
The problem becomes critical when the number of alternatives and experts evaluating
them are large. In this case, it is inappropriate for each expert to carry out a comparison
of all alternatives pairwise. The models described in [4, 6] provide that the expert gives
estimates for some subset of alternatives 𝑋𝑘 from the set 𝑋, where 𝑋𝑘 is the subset of
alternatives chosen by the expert 𝑒𝑘 , 𝑋𝑘 ⊆ 𝑋, 𝑒𝑘 ∈ 𝐸, 𝑘 = 1, ...𝐾.
So, each expert 𝑒𝑘 provides the symbolic estimates for all pairs (𝑥𝑖 , 𝑥𝑗 ), according to
their own scale (LTS), such that 𝑖 = ̸ 𝑗 and 𝑥𝑖 , 𝑥𝑗 ∈ 𝑋𝑘 . Expert’s assessment shows how
much the alternative 𝑥𝑖 is better than the alternative 𝑥𝑗 .
Denote the preference value of the expert for the alternative 𝑥𝑖 with respect to the
alternative 𝑥𝑗 by the variable 𝑝𝑖𝑗 (𝑘).
The obtained values 𝑝𝑖𝑗 (𝑘) form the preference matrix 𝑃 𝑘 =(𝑝𝑖𝑗 (𝑘))𝑖,𝑗=1,...,𝑀 . For
non-estimated alternatives, the symbol “0” is entered in the matrix in the appropriate
position.
In addition, each expert may have their own scale of assessments of the level 𝑡
(LTS): 𝑆(𝑡) = {𝑠1 (𝑡), ..., 𝑠𝑛(𝑡) (𝑡)}, where 𝑠𝑖 (𝑡) is the symbolic assessment of the scale 𝑡,
𝑖 ∈ {1, 𝑛(𝑡)}, 𝑛(𝑡) is the number of assessments, that is, |𝑆(𝑡)| = 𝑛(𝑡). Also, it should be
noted that all assessments are arranged in ascending preferences: 𝑠1 (𝑡) ≺ 𝑠2 (𝑡) ≺ ... ≺
𝑠𝑛(𝑡) (𝑡).
18 ITTMM—2019
Linguistic expression set consists of symbolic assessments, according to which experts
compare alternatives:
𝑆(1) = 𝑠1 (1) : very bad, 𝑠2 (1) : bad, 𝑠3 (1) : normally, 𝑠4 (1) : good, 𝑠5 (1) : very good,
𝑛(1) = 5;
𝑆(2) = 𝑠1 (2) : awful, 𝑠2 (2) : very bad, 𝑠3 (2) : bad, 𝑠4 (2) : normally, 𝑠5 (2) : good,
𝑠6 (2) : very good, 𝑠7 (2) : excellent, 𝑛(2) = 7;
𝑆(3) = 𝑠1 (3) : awful, 𝑠2 (3) : very bad, 𝑠3 (3) : bad, 𝑠4 (3) : below the average, 𝑠5 (3) :
normally, 𝑠6 (3) : above the average, 𝑠7 (3) : good, 𝑠8 (3) : very good, 𝑠9 (3) : excellent,
𝑛(3) = 9.
That structure have solved the problem of misunderstanding between a person and
a computer. To convert the symbolic assessment to the numerical value one of the
well-known methods can be used [17]. We omit the method description since it is not
the subject of our paper.
Accordingly, the matrix 𝑃 𝑘 is modified into a matrix 𝑉 𝑘 =(𝑣𝑖𝑗 (𝑘))𝑖,𝑗=1,...,𝑀 , which
consists of numerical values, such that (𝑠𝑖 (𝑡), 0) → 𝑖, 𝑣𝑖𝑗 (𝑘) = 1, ..., 𝑛(𝑡).
Then, in order to form a matrix of average preference values based on information
obtained from all experts, it is necessary to select a basic linguistic term set (BLTS)
and convert the value from other scales to basic.
Any available LTS can be selected as a BLTS, and by default, the largest scale is
selected, assuming that the scales are ordered by increasing their power, i.e. 𝑆(1) ≺
𝑆(2) ≺ ... ≺ 𝑆(𝑇 ).
To convert estimates from LTS of the level t to BLTS of the level t’, we use the
method described in detail in [6]:
𝛽(𝑡′ ) 𝑛(𝑡′ )
if 𝑆(𝑡) ≺ 𝑆(𝑡′ ) and 𝛽(𝑡) = 𝑛(𝑡) , then 𝐹 : 𝐵(𝑡) → 𝐵(𝑡′ ) mapping is one-to-one and
is determined by the formula (1).
∆−1 (𝑠[𝛽(𝑡)] (𝑡), 𝛽(𝑡) − [𝛽(𝑡)])
𝐹 (𝑠[𝛽(𝑡)] (𝑡), 𝛽(𝑡) − [𝛽(𝑡)]) = ∆( , (1)
𝑛(𝑡)
where 𝛽(𝑡) denotes the shift relative to the nearest integer value in accordance with the
symbolic estimate.
In view of the foregoing the GDM process includes 2 steps.
1) Aggregation of information received from all experts. Individual preferences of
experts are aggregated into the matrix of collective preferences (2). Some problems may
require making a decision in specified restrictions - depending on certain conditions and
situations. To take into account all sorts of restrictions, numerous methods for aggregat-
ing estimates have been developed, for example, aggregating operators. One of these
operators is an ordered weighted average (OWA), as well as its various interpretations.
In our paper we use the simple (i.e. not weighted) average. 𝑃 =(𝑝𝑖𝑗 )𝑖,𝑗=1,...,𝑀 , the
elements of which are calculated as follows
𝐾
∑︀
𝑝𝑖𝑗 (𝑘)
𝑘=1
𝑝𝑖𝑗 = , (2)
𝐾
∑︀
𝑛𝑖𝑗 (𝑘)
𝑘=1
where
⎧
1, if𝑥𝑖 , 𝑥𝑗 ∈ 𝑋𝑘 ,
⎨
𝑛𝑖𝑗 (𝑘) =
⎩0, elsewhere.
2) Exploitation of the information received in step 1. This step forms the final
ranking of alternatives taking into account experts’ preferences.
Chukhno O. et al. 19
We can perform the ranking of alternatives using the mean value between the
operator GDD (Quantifier Guided Dominance Degree) and GNDD ((Quantifier Guided
Non Dominance Degree).
The calculation of the GDD (3) and GNDD (4) is carried out according to following
formulas:
𝑀
∑︁
𝐺𝐷𝐷𝑖 = 𝑝𝑖𝑗 , 𝑖 = 1, ..., 𝑀 ; (3)
𝑗=1
𝑀
∑︁
𝐺𝑁 𝐷𝐷𝑖 = 𝑚𝑎𝑥{𝑝𝑗𝑖 − 𝑝𝑖𝑗 , 1}, 𝑖 = 1, ..., 𝑀 . (4)
𝑗=1
So, we got the Ranking Value. It should be noted, the higher the value, the better
the alternative is considered.
This information is necessary to understand the purpose of developing a chatbot.
We do not give all the formulas of the method of ranking alternatives in the group
decision making process [6], since this work is focused on obtaining expert assessments,
specifically we are developing an environment in which experts better express themselves
in presenting preferences.
2.2. Chatbots
A chatbot is an artificial intelligence software that can imitate a conversation with a
user in natural language using messaging tools, websites or mobile applications [11, 12].
Simply, this is an account that is managed not by people, but by software. The chatbot
works with a less confusing web and mobile application that is easy to install because
there is no need for installation additional packages. Bots are completely different from
human accounts because they do not have online status.
The benefits of such services include efficiency: artificial conversational entity can
combine the steps of complex processes to optimize and automate routine and repetitive
tasks through several simple voice or text requests, reducing execution time and increasing
the efficiency of the task.
Virtual interactive agents can also be deployed in platforms that potential experts
have already used such as Facebook or Twitter, so it is possible to contact with users in
a familiar environment, which increases convenience and makes expert participation in
the process more comfortable and convenient.
Finally, chatbots can simplify and speed up the process of grading or expressing
opinions from a mobile device, browser or any convenient platform. Services maintain
context and control dialogue by dynamically customizing responses based on conversation.
Regardless of the created computer program and platform, human intervention is
crucial in setting up, training and optimizing the system. In order to achieve the desired
results, a combination of different forms of artificial intelligence, such as natural language
processing and machine learning, may be the best choice.
Chatbot is currently described as one of the most advanced and promising ways
of interaction between people and machines. From a technological perspective, such
a program is a natural evolution of the system of answering a question using natural
language processing (NLP) [13, 15].
3. Proposed chatbot system for the group decision making process
In this work, we offer a solution that can be launched inside the Telegram application.
We arrived at decision to work in this cross-platform messenger due to the rich user
interface of the platform bots.
The algorithm of the chatbot in Telegram is quite simple (Fig. 2).
Messages, commands and requests that are sent by users transmit to the software
running on the developers’ servers. Intermediary Telegram Server Webhook handles
20 ITTMM—2019
Figure 2. The architecture model of the chatbot
encryption and provides feedback between the utility and the user. The messenger uses
Webhook technology for authentication and sending notifications about events to your
application. From a programming point of view, it comes down to the work of ordinary
callback functions for processing HTTP requests that will receive data about events,
such as messages received by a chatbot.
Consider the work of the application from a conceptual point of view: when a user
interacts with a chatbot in Telegram, the API (the Application Programming Interface)
sends information about the activity to the code using an HTTP request, and then the
code sends information about how the program should respond (see Fig. 3).
Thus, the Bot API is an intermediary between the Telegram bot and the application
logic. It consists of two main parts: updates and methods. The developer receives
updates reflecting user interaction with the bot. While calling methods are necessary
for a bot to perform predefined actions, such as sending messages to users.
To create a chatbot, it is necessary:
a) Download the Telegram application to computer, phone or to another device.
Telegram is primarily a mobile software tool, but for development purposes, it is
possible to install it on the computer where the code is written. Thus, it is a quickly
way to check the correctness of the work.
b) Get an API key using @BotFather.
After sending the command /newbot, the "name" and the "username" for the bot
should be selected. The name is what other users see in their contact list, and the
username is how they find it.
After completing the steps above, we got an API key. The API key is how Telegram
finds out that the code we wrote is associated with a specific bot. Each chatbot is
assigned a unique API key.
If the name is taken, the bot will ask to enter a new one, if not, it will issue a token
to access the API.
Next, should copy this token to the Telegram driver configuration file. After that
start working with the driver. Another way is to customize bot settings in the bot
@BotFather, including the icon that will be displayed to users.
Chukhno O. et al. 21
Figure 3. A sequence diagram describing the response of the chatbot to an
incoming user message
c) Set up the application configuration (see listing below) and proceed to writing the
code.
Created by the authors of the article bot’s name is GDMbot. The purpose of
developing this virtual assistant is to facilitate the process of interaction of the expert with
the system. The bot configuration code with the user name @GroupDecisionMakingBot
is shown below.
We want the expert to feel as comfortable as possible in the rating process. Therefore,
when writing a chatbot program, the authors planned that in order to communicate
with the bot, the user would not have to type the message text, since the bot interface
would provide a set of user buttons. To do this, we implement a virtual menu.
After all the code is ready, ought to register the hosting and add the domain name to
it, transfer the script to the database and start the chatbot. At this stage, it is needed
to have a web host. It is most convenient to create a separate subdomain for the bot -
for example, bot.example.com.
4. Experimental results
The authors of this very article have developed the @GroupDecisionMakingBot for
the group decision making process. Computer program of chatbot will help experts
to express their opinions and set ratings in any place and at any time, using a mobile
device, tablet or personal computer.
Developing the concept described in [6], we propose to automate the entire process
of group decision making. According to the method of this article, a software tool has
Chukhno O. et al. 23
Figure 4. Chatbot interface: Chatbot response snapshots
already been developed that defines ranked lists of alternatives. However, the bulk of the
time was spent communicating with experts and receiving assessments. We also noticed
that the experts were tired in the process of providing preferences, which adversely
affected the reliability of the data obtained. Therefore, we consider this work to be the
completion of the full automation of the entire group decision making process.
To pass a survey and provide estimates to the system, each expert needs to find our
chatbot in Telegram by its name, more specifically, call @GroupDecisionMakingBot.
After the greeting, the user will be asked to participate in the survey and compare
several of the 27 social networks pairwise. Experts can manage their actions using the
“Forward” / “Back” buttons, so the expert may not be afraid of random errors, since it
is always possible to correct any answers.
The second step is the definition of the user rating scale: the system offers 3 options,
discussed in detail in [4, 6].
The final step is a pairwise comparison of the selected alternatives: the bot automati-
cally displays a couple of alternatives and user response options, according to the rating
scale. Note that the expert does not need to remember which social networks he chose,
because the interactive system will independently request all the necessary information.
Finally, the bot sends to the creator of the application the results obtained from the
experts. Further alternatives can be ranked using a previously developed software tool
[6].
5. Features of chatbot use in the group decision making process
The @GroupDecisionMakingBot has the following advantages:
– simple and interactive real-time chat system;
– an effective way for experts to interact with the system;
– providing preferences anywhere and anytime;
– ability to send survey results to the administrator;
– work in cross-platform devices;
– easy integration and upgrade of the application;
24 ITTMM—2019
– unlimited members.
The advantages listed above make the chatbot unique from the point of view of its
use in the group decision making process, since in this area a set of such characteristics
is implemented for the first time.
6. Conclusions
The novelty of this paper consists in that a chatbot is used in the group decision
making process. Moreover it becomes possible for an unlimited number of experts
to participate, and also provides comfortable conditions for their work according to
advanced technologies.
In the future, it is planned to develop a software tool that can recognize speech and
the image of experts to provide an environment in which experts will communicate with
a machine like with people.
Acknowledgments
The publication has been prepared with the support of the “RUDN University
Program 5-100” and funded by RFBR according to the research projects № 18-07-00576
and № 18-00-01555 (18-00-01685).
References
1. Scott, J. Social network analysis. Sage, 2017.
2. J. Kacprzyk, Group decision making with a fuzzy linguistic majority, Fuzzy sets
and systems 18, 105–118, 1986.
3. E. Herrera-Viedma, F. J. Cabrerizo, J. Kacprzyk, W. Pedrycz, A Review of Soft
Consensus Models in Fuzzy Environment, Information Fusion, 17:4–13, 2014.
4. E. Herrera-Viedma,F. J. Cabrerizo, F. Chiclana, J. Wu, M. J. Cobo, K. Samuylov,
Consensus in Group Decision Making and Social Networks, Studies in Informatics
and Control 26:3, 259-268, 2017.
5. J. A. Morente-Molinera, I. J. P´erez, M. R.Ure˜na, E. Herrera-Viedma, On multi-
granular fuzzy linguistic modeling in group decision making problems: a systematic
review and future trends, KnowledgeBased Systems 74, 49–60, 2015.
6. N. Chukhno, O. Chukhno, A. Gaidamaka, K. Samuylov, E. Herrera-Viedma, A New
Ranking Method of Alternatives for Group Decision Making in Social Networks,
10th International Congress on Ultra-Modern Telecommunications and Control
Systems and Workshops (ICUMT), 2018.
7. V. Begishev, R. Kovalchukov, A. Samuylov, A. Ometov, D. Moltchanov, Y.
Gaidamaka, S. Andreev, An analytical approach to SINR estimation in adjacent
rectangular cells, Lecture Notes in Computer Science (including subseries Lecture
Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9247,
446-458, 2015.
8. A. Samuylov, D. Moltchanov, Y. Gaidamaka, S. Andreev, Y. Koucheryavy, Random
Triangle: A Baseline Model for Interference Analysis in Heterogeneous Networks,
IEEE Transactions on Vehicular Technology, vol. 65, no. 8, art. no. 7275184, 6778-
6782, 2016.
9. K.E. Samouylov, P.O. Abaev, Y.V. Gaidamaka, A.V. Pechinkin, R.V. Razumchik,
Analytical modelling and simulation for performance evaluation of sip server with
hysteretic overload control, Proceedings - 28th European Conference on Modelling
and Simulation, ECMS 2014, 603-609, 2014.
10. Y. Gaidamaka, A. Pechinkin, R. Razumchik, K. Samouylov, E. Sopin, Analysis of
an MG1R queue with batch arrivals and two hysteretic overload control policies,
International Journal of Applied Mathematics and Computer Science, 24 (3), 519-534,
2014.
Chukhno O. et al. 25
11. S. A. Abdul-Kader, J. Woods, Survey on Chatbot Design Techniques in Speech
Conversation Systems, International Journal of Advanced Computer Science and
Applications, 6 (7), 2015.
12. M. Dahiya, “A tool of conversation: Chatbot, International Journal of Computer
Sciences and Engineering, vol. 5, no. 5, 158–161, 2017.
13. T. Kluwer, From chatbots to dialog systems, in Conversational agents and natural
language interaction: Techniques and effective practices. IGI Global, pp. 1–22, 2011.
14. S. Sayed, R. Jain, B. Lokhandwala, F. Barodawala and M. Rajkotwala, Android
based Chat-Bot, International Journal of Computer Applications, Vol. 137, No. 10,
29-32 , 2016.
15. F. A. Mikic Fonte, M. L. Nistal, J. C. Burguillo Rial, and M. C. Rodríguez, NLAST:
A natural language assistant for students, IEEE Global Engineering Education
Conference (EDUCON), 709-713, 2016.
16. M. N. Kumar, P. C. L. Chandar, A. V. Prasad and K. Sumangali, Android based
educational Chatbot for visually impaired people, 2016 IEEE International Confer-
ence on Computational Intelligence and Computing Research (ICCIC), Chennai,
1-4, 2016.
17. V. Torra, Hesitant fuzzy sets, International Journal of Intelligent Systems, 25(6),
529-539, 2010.