Chatbot for Education System

Aishwarya Hajare, Priyanka Bhosale, Rasika Nanaware, Guruswami Hiremath

[email protected], [email protected], [email protected], [email protected]
B.E. Computer Dept.,
AISSMS IOIT, Kennedy Road, Pune

ABSTRACT develop many applications one of that is

The purpose of this paper is t o develop an automated system mentioned in this paper is chat bot for the
which gives reply to user query on behalf of human for education system. Though chat but has a variety of
education system. It can give answer to each and every query fields like marketing, medical, education, banking
asked by end user. Existing Chabot such as Facebook chat bot, and finance. We will see related work in each
WeChat, Natasha from Hike, Operator, etc. were giving reply field.
from its local database. But our approach is to focus on
local database as well as web database and also to make
Kyo-Joong oh & Dong Kun Lee [1]
system scalable, user friendly, highly interactive. Various
techniques such as neural network (RNN), NLP, pattern Medical field chatbot is useful for providing free
matching data mining algorithms are used in this paper to counseling to the patient, suggesting various
enhance the performance of the system medicines, etc. Even without going to hospital or
Keywords visiting with the doctor. Kyo-Joong oh & Dong
Pattern Matching, RNN, NLP, Web mining, Deep learning, Kun Lee in [1] have proposed a chat bot psychiatric
Response generation. counseling in mental health care service. Emotion
recognition is used to implement this chatbot.
Because of this, the patient is not needed at all to go
to hospital. Free counseling is provided at the user’s
1. INTRODUCTION location.
The development in the information technology and
communication has made artificial intelligent systems more
complex. The AI systems are approaching human activities Nikita Hatwar [2]
such as taking a decision at a particular moment, performing Nikita Hatwar in [2] proposed another interesting
day to day tasks. In an artificial intelligent field, there are some chatbot for marketing field named as AI based chat
hybrid methods and adaptive methods available which are bot. In this paper, this chat bot is used for providing
making systems more complex. Not only that, but also there is guidance to the visitor of the mall, provides
a hybrid combination of natural language processing and
navigation and according to latest discount going in
intelligent systems. These systems can learn themselves and
the shop, it was guiding recommendation system is
renew their knowledge by reading all electronics articles
implemented in this project according to the rating
available on the internet Human as a user can ask to the
we can advise people to go to a particular shop.
systems like usually ask to another human. These systems are
often known as internet answering engines. In addition to the This feature can be extended further.
internet answering engines, currently many applications are
introduced such as chatter-robot or known as Chabot which is Prof. Yu Wo and Gongxiao Wang [3]
often aimed for giving automatic reply or just for In the marketing field, the user has lots of
entertainment. This application's work is very simple, because questions related to projects whose answer is not
the knowledge is already programmed in advance. Few of present even in local database, to overcome this
the methods used in this application are pattern-matching, problem Prof. Yu Wo and Gongxiao Wang in [3]
natural language processing, data mining. The Chabot would presented paper automatic chat bot knowledge
match the input sentence from the speaker or user with that
acquisition from online via rough set and ensemble
pattern existed in the knowledge base. Each pattern is then
learning. The concept of ensemble learning is used
compared with the knowledge of Chabot. This knowledge has
here. The which constructs classification results of
been taken from various sources. Rest of the paper is organized
the learner to get the final result. Multiple rough
as follows. Section2 describes the existing systems section 3
describes proposed system, section 4 describes text set, classifier is constructed and trained first, then
Classification used in Chat Bot and section 5 describes the all replies are classified with these classifieds. The
result followed by conclusion and future scope.. final results are drawn from voting to the output of
these classifiers and finally it is selected as
knowledge database. The disadvantage of this
system is that not all replies are related to root
2.SURVEY DETAILS message different forum have different styles and
By using artificial Intelligence field, we can formats.

Volume: 3 Issue: 2 April - 2018 52

Naveen Kumar and Linga chandar P.C [4]
The education system also has chat, but Naveen
Kumar and Linga chandar P.C in [4] have
proposed an android based educational chat bot for
visually impaired people. Voice processing,
symbolic reduction and keyword detection
methodologies are used. In case of noisy
environment application is not recommended.
Application should be scalable, so even if
thousands of queries are asked at a time. Then also
system is able to answer all the queries.

Juan Carlos, et al [5]

Juan Carlos, et al [5] proposed system that is
chatting with an Arduino platform through the
telegram boy. The proposed system is connecting
telegram and Arduino platform using telegram boy.
This system allows people to communicate with
machine user, where this machine user can operate
simple and complex prototype developed by the
same person. To implement this system, telegram
apes are used. REST protocol is used for the
integration of both.

Jacques Garcia Faust Abraham [6]

For Social networking field Jacques Garcia Faust Abraham
have proposed and developed a Chabot in a social network in
[6], This paper describes how to implement chat bot on twitter
for entertainment and viral advertising using database and
simple algorithms. Message reception, message passing,
me s s a ge processing techniques is used. With this system a
person can earn more followers every day without any human
intervention.

Wenzeng Zhu [7]

Fig: Proposed System Architecture
In social networking security is a main concern to
implement this functionality, Wenzeng Zhu in [7]
proposed a system that is internet security
protection for IRC based Botnet many works
contain infected data which can harm our system.
So in this paper, they proposed a noble paper based
on IRC analysis and detection to protect our
computers. IRC servers, used instead of C&C
server. So the infected computers link to the IRC
server. This method can protect the infected
computers not to return Bots controlled by the Bot
Master. The only problem with this is that we need
to detect all the infected computers and help them
to uninstall bot code. To return securely, but the
problem is that some boot code is hard to uninstall.

3.Proposed System Architecture

Volume: 3 Issue: 2 April - 2018 53

4.Flowchart of the proposed system: 5.Generated OTP is sent to the user through
email a n d mobile number. When the user
receives the OTP, the user has to enter that OTP.
6.On success, the browser gives a refresh token server take this
token for the user and call refresh token. Here verification
process will end.
7. User i n p u t t h e queries in the he chats b o x ,
this query(string) is encrypted and sent to chat
server.
8.Chat server decrypts the per user queries and look for the
tags and keywords through Clustering and Reduction
techniques.
9.Create tag and keywords weights for each query.
10. Match tags and keywords with their weights to
find top probabilities.
11. If the keyword probabilities are low (not
available in dB) then try to find answers for
queries by data mining techniques.
12. If a result is found, then return () function
will return the answer to user, else send the
message as an admin will contact with you by
email.
13. Stop.

5. Text Classification used in Chat Bot

5.1 Pattern Matching

Early Chabot’s used pattern matching [8-10]to
classify text and produce a response. This is often
referred to as “brute force” as the author of the
system needs to describe every pattern for which
there is a response.

A standard structure for these patterns is

“AIML” (artificial intelligence markup
language).

<aiml version = "1.0.1" encoding = "UTF-8"?>

<category>
<pattern>How college campus is? </pattern>
<template>Collage campus is very huge. Then also I suggest
you to google it.
Fig: System Flowchart </template>
</category>
Working of the Proposed System: <category>
<pattern> What is the name of the Principal? </pattern>
1.Start <template>Dr. P.B. Mane</template>
</category>
2.The user opens our college web page inside his </aiml>
browser, automatically Chabot panel will be Pop-Up. The machine then produces:
Human:How college campus is?
3.The user must have to enter mobile number and email id. Robot: Collage campus is very huge. Then also I suggest you
4.Entered users data is sent to OTP server and OTP is to google it.The working of Pattern matching is totally based
generated. on main keywords. The keywords like placement, cut off, etc.

Volume: 3 Issue: 2 April - 2018 54

are extracted from the query and are given for processing.
5.3 Neural Network:
5.2 Algorithms Used
Naive Bayes [11-14] is a simple technique for constructing
The neural network model called RNN
classifiers: models that assign class labels to problem instances,
represented as vectors of feature values, where the class labels Encoder-Decoder that consists of t w o
are drawn from some finite set. It is not a single algorithm for recurrent neural networks (RNN). One RNN
training such classifiers, but a family of algorithms based on a encodes a sequence of symbols into a fixed-
common principle: all naive Bayes classifiers assume that the length vector representation, and the other
value of a particular feature is independent of the value of any decodes the representation into another
other feature, given the class variable. For example, a fruit may sequence of symbols. The encoder and
be considered to be an apple if it is red, round, and about 10
cm in diameter. A naive Bayes classifier considers each of
decoder of the proposed model are jointly
these features to contribute independently to the probability trained to maximize the conditional
that this fruit is an apple, regardless of any possible probability of a target sequence given a source
correlations between the color, rounded, and diameter features. sequence. The performance of statistical
For some types of probability models, naive Bayes classifiers machine translation system is empirically
can be trained very efficiently in a supervised learning setting. found to improve by using the conditional
One can work with the naive Bayes model without accepting
probabilities of phrase pairs computed by the
Bayesian probability or using any Bayesian methods.
RNN Encoder-Decoder as an additional
feature in the existing log-linear model.
Qualitatively, we show that the proposed
model learns a semantically and
syntactically meaningful representation of
linguistic phrases.

An advantage of naive Bayes is that it only requires a

small number of training data to estimate the parameters
necessary for classification
A sample training set:
Class: Greeting
“Hi there!”
“How are you?”
“Hello”
Class: Placement
“Which companies are approaching college?”
“What is the salary range?”
“Is TCS, Tech Mahindra to the college?”
Class: Staff Fig. RNN
“For computer department hoe the staff is?”
“What is the salary range?” The scenario of neural network of chat both
“Education of staff?” for the education system is drawn and
Let us classify a few simple input sentences: explained in figure 1.3. As input indicates
Input: “Hi there” the number of inputs from the user.
Term:” Hi (No matches)” Consider the scenario for Input1. If the user
Term: “there (Class: Greeting)” asked about placement. It is shown with green
Classification: greeting(score=1) circle. Blue c i r c l e (Hidden Layer)
Input: “Which companies are visiting to the college?” indicates n um be r o f c l a s s e s s u c h as
Term: “Companies” (Class: Placement (2)) placement of IT department, p l a c e m e n t
Term: “visiting” (Class: Placement (2)) of the computer department, placement of
Classification: Placement (Score (4)) electronics department, e t c . The processing
By using an equation, we a r e l ook i n g for is done at hidden layer based on the classes.
word matches given some sample sentences And the output layer generates the output.
for each class, and we avoid having to identify
every pattern.

Volume: 3 Issue: 2 April - 2018 55

6. Result:

(a)User credentials

(c) Communication
As shown in fig (a) user enters mobile
number and Email id for verification, in fig
(b) user enter received OTP if it is correct
then communication start which is shown in
fig(c). In this way system will produce output.
Only the appropriate keywords are extracted
from the given query and using data mining
algorithm data will be mined from the
knowledge based. Sentence will be formed
and presented in front of user.

(b)Generate OTP

7. CONCLUSION AND FUTURE SCOPE: 8. REFERENCES

The proposed Chabot is an amalgamation of three
techniques such as pattern matching, neural
network and naive bays algorithm. Because of this, [1] K. J. Oh, D. Lee, B. Ko and H. J. Choi, "A
end user will not come to know whether human is Chabot for Psychiatric Counseling in Mental
giving reply or a system. Queries which are not Health Care Service Based on Emotional
answered by Chabot are answered by human Dialogue Analysis and Sentence Generation,"
through E-mail. This system can be extended for 2017 18th IEEE International Conference on
banking sector by providing an additional layer of
Mobile Data Management (MDM), Daejeon,
security. In order to decrease query processing
2017, pp. 371-375
time and serve results within a short period of time
advanced query processing algorithm can be used.

Volume: 3 Issue: 2 April - 2018 56

[2] Prof. Nikita Hatwar “chat bot for marketing International Conference on Image, Vision and Computing
field named as AI based chat bot.”In 2016 (ICIVC), Chengdu, 2017, pp. 122-127

[3] Y. Wu, G. Wang, W. Li and Z. Li, "Automatic Chatbot [9] M. U. Nisar, S. Voghoei and L. Ramaswamy, "Caching for
Knowledge Acquisition from Online Forum via Rough Set and Pattern MatchingQueries in Time Evolving Graphs:
Ensemble Learning,"2008 IFIP International Conference on Challenges and Approaches," 2017 IEEE
Network and Parallel Computing, Shanghai, 2008, pp. 242-
37th International Conference on Distributed
246.
Computing Systems (ICDCS), Atlanta, GA, 2017,
pp. 2352-2357.
[4] M. N. Kumar, P. C. L. Chandar, A. V. Prasad and K.
Sumangali, "Android based educational Chatbot for visually
impaired people," 2016 IEEE International Conference on [10] D. G. Bailey, "Identification of Trax threats using pattern
Computational Intelligence and Computing Research (ICCIC), matching," 2016 International Conference on Field-
Chennai, 2016, pp. 1-4. Programmable Technology (FPT), Xi'an,2016, pp. 319-322.

[5] J. C. De Oliveira, D. H. Santos and M. P. [11] İ. Ü. Oğul, C. Özcan and Ö. Hakdağlı, "Fast text
Neto, "Chatting with Arduino platform through classification with Naive Bayes method on Apache Spark,"
Telegram Bot," 2016 IEEE International 2017 25th Signal Processing and Communications
Symposium on Consumer Electronics (ISCE), Applications Conference (SIU), Antalya, 2017, pp. 1-4.
Sao Paulo, 2016, pp. 131-132
[12] N. Sharma and M. Singh, "Modifying Naive Bayes
[6] Randall Wald, Taghi M. Khoshgoftaar, Amri classifier for multinomial text classification," 2016
Napolitano, Chris Sumner, "Predicting International Conference on Recent Advances and Innovations
susceptibility to social bots on Twitter", in Engineering (ICRAIE), Jaipur, 2016, pp. 1-7.
Information Reuse and Integration (IRI) 2013
IEEE 14th International Conference on, pp. 6- [13] H. Parveen and S. Pandey, "Sentiment analysis on Twitter
13,2013. Data-set using Naive Bayes algorithm," 2016 2nd
International Conference on Applied and Theoretical
[7] W. Zhu and C. Lee, "Internet security protection for IRC- Computing and Communication Technology (iCATccT),
based botnet,"2015 IEEE 5th International Conference on Bangalore, 2016, pp. 416-419.
Electronic Information and Emergency Communication,
Beijing, 2015, pp. 63-66. [14 ]W. Feng, J. Sun, L. Zhang, C. Cao and Q. Yang, "A
support vector machinebased naive Bayes algorithm for spam
filtering," 2016 IEEE 35th International Performance
[8] Qiuze Yu, Shunxin Min, Bo Pang and Yan Zhang, Computing and Communications Conference (IPCCC), Las
"Simultaneously Multiple-object Pattern Matching based on Vegas, NV, 2016, pp. 1-8
Multi-swarms Particle Swarm Optimization," 2017 2nd

Volume: 3 Issue: 2 April - 2018 57