University Chatbot using Artificial Intelligence

Markup Language
Naing Naing Khin Khin Mar Soe
Natural Language Processing Lab Natural Language Processing Lab
University of Computer Studies,Yangon University of Computer Studies,Yangon
Yangon, Myanmar Yangon, Myanmar
naingkhin@ucsy.edu.mm khinmarsoe@ucsy.edu.mm

Abstract—Chatbots are conversational systems that can do tending to buy things online. S. Hussain, O. A. Sianaki, and
chat interactions with human automatically. It is developed to N. Ababne[4] discussed chatbots classification, design
be virtual assistant, making entertainment for people, helping
for answering the questions, getting driving directions, serving techniques used in earlier chatbots and modern ones, and
as human partner in smart homes etc. Most of the chatbots how the main categories of chatbots can handle conversation
utilize the algorithms of artificial intelligence (AI) in order to get context. They presented with the emergence of new
the required responses. In this paper, we provide the design of a
technologies intelligent systems have emerged using
University Chatbot that provides an efficient and accurate
answer for any user questions about university information. complex knowledge-based models. B. R. Ranoliya, N.
This is the first University Chatbot for inquiring about school Raghuwanshi and S. Singh[5] provided the university related
information in Myanmar Language based on Artificial FAQs for students by using the Artificial Intelligence and
Intelligence Markup Language and uses Pandorabots as the
interpreter.
Latent Semantic Analysis. They implemented to solve the
academic needs of the visitors for Manipal University. A.
Keywords—AIML, Natural Language Processsing, Mondal, M. Dey, D. Das, S. Nagpal, K. Garda[9] focused on
Pandorabots, Pattern Matching, Response Generation
the design of texual communication in educational domain.
I. INTRODUCTION They developed and processed the accurate chatbot by using
Conversational agents become essential by interacting of random forest algorithm. R. Sharma, M. Patel[10] presented
machines with the desired users to provide natural language the review of design techniques on chatbot in speech
interfaces. So, the role of chatbots in the information conversation systems. They discussed the performance and
technology and communication is widely in used. Many usability of many chatbots in our everyday lives.
chatbots are created day by day through marketing, medical,
education and banking. Chatbot is also a user assistant III. HUMAN COMPUTER INTERACTION IN CHATBOTS
substance that is intended to produce a communication with
human through their regular language. In educational system, Human Computer Interaction is a communication field of
it is essential for teaching, learning and searching the desired study focusing on the design of computer technology and the
information for a specific area. The obvious factor that leads interaction between human and computers. Conversation
us one step closer to living in our fantastic world is that it system between a human and a computer is either chatting by
knows our messages and can respond to us. The bot would
match the input sentence from the user with that pattern typing text or speech dialogue using the voice. Thus,
existed in the knowledge base. This system is simple interaction with natural language is a feasible option for
Myanmar chatbot using AIML but it can answer the necessary connecting machine agents and human users. The chatbot
information for the users. There are many chat engines with popularity has brought the new feeds for HCI as it has
different methods and can perform chatting. Some famous changed the pattern of human interaction. Human Computer
chatbots are SimSimi, Mitsuku, A.L.I.C.E, and now the
machine learning chatbots like Siri, Alexa, Cortana and so on. Interaction may need to consider for chatbots as the main
Although modern chatbots apply the power of artificial object of design, focus on services than user interfaces, and
intelligence to answer complicated questions, they still need design for interaction in networks of human and machine
some improvements for low resource languages. actors[13]. The main parts which include human computer
II. LITERATURE REVIEW interaction in conversation systems design are (a) the
techniques used to produce keywords, (b) pattern matching
B. A. Shawar and E. Atwell [2] described a system to
techniques used inside the chatbot and (c) the type of
access Arabic language information using chatbot without
response. So, HCI is considered taking on human-chatbot
sophisticated natural language processing or logical
interaction design as an area of research and practice.
inference. They showed this work properly with English and
European languages and how the same chatbot will react
in terms of Arabic FAQs. S. A. Prasetya, A. Erwin[3] aimed Make a
Conversation
to found right AIML interpreter which can be used in Chatbot Response
Keywords
Indonesian language E-Commerce website. The system is
built by using Artificial Intelligence Markup Language
Fig. 1. Human Computer Interaction
(AIML) and Pandorabots as the interpreter for the customers

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

 IV. ARTIFICIAL INTELLIGENCE MARKUP LANGUAGE sentences can be carefully checked out from the user input.
AIML is an XML based markup language for specifying Recursive categories involve many applications: (i) symbolic
chatbot content. It was created by the ALICE bot free reduction which reduces the complex grammatical forms to
software community in 1995-2000 for the people to input simpler ones; (ii) divide and conquer category splits an input
dialogue pattern knowledge into chatbots based on the into two or more subparts and add the responses to one; (iii)
ALICE free software technology. An AIML Interpreter is synonyms resolution is possible to appear different words
able to load and run the bot, then, it provides the bot’s with the same meanings depending on the consisting text; (iv)
responses in a chat session with a user. AIML consists of data keyword detection is possible to find the same response
objects called AIML objects, which are made up of units when a definite keyword is found in the user input [6]. These
called topics and categories. The topic is called an optional are some examples of different AIML categories.
top-level element, it has a name and a set of categories related
to that topic. Categories are the basic units of knowledge in (i) symbolic reduction
AIML. One category is a rule for matching an input and
converting to an output, and consists of a pattern, which <category>
represents the user input, and a template, which responses <pattern>တခ ယသ က သလ </pattern>
the answer. The AIML pattern is simple and consists of <template>က"နပ #တ သပ$မ % ဟ ဌ န မ ဌ နမ#
words, spaces, and the wildcard symbols _ and *. 'ဖစပ တယ </template></category>
<category>
<?xml version="1.0" encoding="UTF-8"?> <pattern>တခ ယမ"န က သလ </pattern>
<aiml version="2.0"> <template>သတသပ$မ % ဟ ဌ န မ ဌ နမ# 'ဖစပ တယ
<!—AIML code goes here --> </template></category>
</aiml> <category>
<pattern>သင * က သလ </pattern>
<template><srai><star/> က သလ </srai>
A. AIML Categories
</template></category>
There are three AIML types: (a) atomic categories, (b)
default categories, and (c) recursive categories[1].
(a) Atomic categories are those with patterns that do not
have wildcard symbols, _ and *. (ii) divide and conquer

<category> <category>
<pattern>မဂလ ပ </pattern> <pattern> က ဇ* ပ ရင</pattern>
<template>ဟတက မဂလ ပ ရင</template> <template>ဟတက 'ပနလည တ"ဆ%ဖ မ- လငပ တယ
</category> </template> </category>
<category>
<pattern> က ဇ* ပ *</pattern>
(b) Default categories include wildcard symbols * or _. <template> <srai> က ဇ* ပ ရင </srai>
</template></category>
<category>
<pattern> က င ခ န် က ဘယအခ နမ * </pattern>
<template> က င ခ န က မနက ၉ န ရမ စပ တယ (iii) synonyms resolution
</template>
</category> <category>
<pattern> က င လပစ သခ ငလပ </pattern>
<template>အမတ-၄ လမ မ0က ၊ ရ2'ပညသ 3မ့ု၊
ရနကနတင 'ဖစပ တယ</template>
For such situation, if the user enters က င ခန က
</category>
ဘယအခ နမ စတ လ then the AIML class will search until the <category>
wild symbol (*) and if there is a match it will accompany <pattern> က င က ဘယမ ရတ လ</pattern>
response. <template><srai> က င လပစ သခ ငလပ </srai>
(c) Recursive categories are the categories with </template> </category>
templates <srai> and <sr> tags, which represent recursive
artificial intelligence and symbolic reduction. Applying a
combination of wild cards and srai, the stop words of the
 (iv) keyword detection A. Knowledge Base
Artificial Intelligence Markup Language is a well-known
<category> XML derived language to build chatbot knowledge base.
<pattern>သငယ*'ခင </pattern> Users' frequent asked question sets are defined semantically
<template> သငယ*'ခင သည 'ပ င လ'ခင ဆသ the knowledge domain given to the chatbot. The questions are
ဦ တည စ သ လပငန စဥ် 'ဖစသည</template> available from the university academic center related to nine
</category> topics and manually collect from the teachers, students and
<category> their parents that they want to ask about the university. We
<pattern> _ သငယ*'ခင </pattern> have used 970 question-answer pairs as data distribution. A
<template><srai>သငယ*'ခင </srai></template> well-designed knowledge base can positively impact the
</category> effectiveness of chatbots that will improve the interaction
<category>
between users.
<pattern>သငယ*'ခင *</pattern>
<template><srai>သငယ*'ခင </srai></template> B. Workflow of the System
</category>
<category>
<pattern>_ သငယ*'ခင *</pattern> Start
<template><srai>သငယ*'ခင </srai></template>
</category>

Accept User
Query
V. PROPOSED SYSTEM
We have implemented a Myanmar interactive chatbot for
university frequently asked questions. AIML is defined with
general inquiries and messages which are replied by applying Word Segmentation
AIML formats. According to the Artificial Intelligence
Markup Language, we have used different AIML tags to get
the user required information from the bot. Fetch Query to
Pandorabots’ Server
TABLE I. AIML TAGS USED IN SYSTEM

No. Tags used for AIML Categories Knowledge

Base
1. <topic> </topic> Pattern Matching using the
2. <category></category> Pandorabots API

3. <pattern></pattern>
4. <template> </template>
5. <srai></srai> Response of
Query
6. <random> </random> with <li> </li>
7. <set> </set>
8. <get> </get> End
9. <that> </that>
10. <think> </think> Fig. 2. Flow of the System
11. <condition> </condition>
VI. IMPLEMENTATION OF THE PROPOSED CHATBOT
Rule-based chatbot contains a faster time-to-relevance,
The system operation is divided in three steps. In the first delivering a faster impact on user interaction. This chatbot is
step, the question is entered by the user. In the second step, the one of rule-based chatbot and developed on AIML language
system performs word processing actions to match the user's for the University of Computer Studies, Yangon. We have
input to a pre-defined format and do the pattern matching purposed this system to have a support for university routine.
between user input and the Knowledge Base. Finally, the All the questions files need to be uploaded to Pandorabots
answer is presented to the user in the third step. server. The files include the university related questions and
information that the students, teachers and parents frequently
asked. The number of question-answer pairs in the system that
are utilized for different topics and type of categories are as
given:

TABLE II. THE TOPICS AND NUMBER OF QA PAIRS USED IN THE

SYSTEM

Topics Atomic Default Recursive

Greeting 25 12 23
Location & 13 15 7
Address
Academic 172 200 61
Brief 18 20 15
History
Conference 39 35 11
Faculties & 50 25 22
Staff
Library 40 20 10
Research & 34 38 20
Lab
Alumni 12 25 8
Fig. 3. Sample Outputs of the System

The service requires internet connection to access the VII. TESTING AND EVALUATION
system. The users can interact with the chatbot in every time Testing can be made to measure the quality of chatbot. The
if there is a connection. The user needs to input the questions steps included to conduct the chatbot experiments are (i)
as Myanmar Language. The segmentation is done by using the getting the overviews of questions that can be asked, (ii)
UCSY word segmentor (http://www.nlpresearch- inquiring the user questions related to the nine topics show in
ucsy.edu.mm/NLP_UCSY/wordsegmentation.html). The Table 2, (iii) asking the questions to get the feedback about the
input from the user is normalized and processed on the system, (iv) the experimental analysis whether they have
Pandorabots server. The AIML files, which are separated into correct or wrong responses [12].
several categories and the chatbot's knowledge is uploaded
into the server. After pattern matching, the inquired user can We have made user testing with the closed user group by
ask the university related questions about academic services students, parents and staff. The system still needs
and activities. These are some sample results of our system: improvement due to some patterns that mismatch with the
chatbot knowledge. We also have adopted the dialog
efficiency matrix to evaluate the chatbot.
A. Dialogue Efficiency Metric
The proposed system is measured the efficiency with four
sample dialogues in terms of atomic categories, default
categories, recursive categories. Adopted learning
mechanism is used to see the ability to find answers for user
as shown in table III.

TABLE III. TESTING IN FOUR DIALOGUES TYPES

Matching Type D1 D2 D3 D4
Atomic Categories 8 13 20 10
Default Categories 25 9 7 19
Recursive Categories 14 10 15 20
Total 47 32 42 49
Mismatch Pairs 3 2 3 6

In the above table, we have tested the system with four

dialogues as D1, D2, D3 and D4 for three matching
categories. The bot can answer most of user questions
correctly. We have also found some miss-match questions
and answers as in figure 4. The error may occur by the user
input or may be the chatbot knowledge. When we have
multiple categories with the same pattern, AIML interpreter
chooses categories from bottom to top within each individual VIII. CONCLUSION
files. Unexpected bot responses can cause due to
Chatbots can interact with people in effective ways. There
normalization error. This can be solved by the substitution are many chatbots in English and other languages by using
file in the system. We also need to prepare more data for different algorithms and models but there is little chatbot
chatbot's knowledge. using Myanmar language. This is one of the University
Chatbots using Myanmar Language to fulfill the information
gaps between the university and its related users. Now, we
have implemented a chatbot for the University of Computer
Studies, Yangon. This is simple chatbot using Artificial
Intelligence Markup Language and implemented on the
Pandorabots server. The user can ask the useful questions
about the university related the academic sectors through
chatbot. The bot will help people to save time and get the
information every time. We still need some improvements for
the bot and we will develop this by using machine learning
techniques in coming jobs.

REFERENCES
[1] AIML Foundation, June 2018.
[2] B. A. Shawar, E. Atwell, "Arabic question-answering via instance
based learning from an FAQ corpus", 2009.
[3] S. A. Prasetya, A. Erwin, M. Galinium, " IMPLEMENTING
INDONESIAN LANGUAGE CHATBOT FOR ECOMMERCE SITE
USING ARTIFICIAL INTELLIGENCE MARKUP LANGUAGE
(AIML)", 2018.
[4] S. Hussain, O. A. Sianaki, N. Ababneh, "A Survey on Conversational
Agents/Chatbots Classification and Design Techniques", 2019.
Fig. 4. Mismatch Question Example [5] B. R. Ranoliya, N.Raghuwanshi, S. Singh, "Chatbot for university
related FAQs," 2017 International Conference on Advances in
The frequency of matching type in each dialogue Computing, Communications and Informatics (ICACCI), Udupi, 2017,
pp. 1525-1530.
generated between user and chatbot was calculated in figure [6] AIML - Sets and Maps in AIML 2.0,
5. These absolute frequencies are normalized to relative https://docs.google.com/document/d/1DWHiOOcda58CflDZ0Wsm1C
gP3Es6dpicb4MBbbpwzEk/pub
probabilities.
[7] B. A. Shawar, E. Atwell ,"Chatbots: Are they Really Useful?", 2007.
[8] A. Hajare, P. Bhosale, R. Nanaware, G. Hiremath, "Chatbot for
60 Education System", Vol.3, Issue 2, April 2018.
[9] A. Mondal, M. Dey, D. Das, S. Nagpal, K. Garda, "Chatbot: An
50 automated conversation system for the educational domain", IEEE,
2018.
40 [10] R. Sharma, M. Patel, "Review on Chatbot Design Techniques in Speech
Conversation Systems", Vol.5, Issue 9, September 2018.
30 [11] A. S. Lokman, J. M. Zain, "Chatbot Enchanced Algorithms: A Case
Study on Implementation in Bahasa Malaysia Human Language",
20 2010, pp.31-44.
[12] B. A. Shawar, "Different measurements metrics to evaluate a chatbot
10 system", 19 May 2014.
0 [13] A. Følstad, P. B. Brandtzaeg, "Chatbots and the new world of HCI", 15
April 2018.
Atomic Default Recursive Total Mismatch
Categories Categories Categories Pairs [14] Giovanni De Gasperis, Isabella Chiari, "AIML Knowledge Base
Construction from Text Corpora", 5 June 2014.
D1 D2 D3 D4

Fig. 5. Dialogue Efficiency of the Tested Dialogues