Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 5

Real Time Sentiment Analysis of

Student’s Feedback
ABSTRACT

Educational Data Mining (EDM) plays vital role in the improvement of education by keeping a check on the performance of
the student in his/her studies and by trying to understand learning of the students. Conducting feedbacks from students at the
end of the year/semester has an issue and disadvantage, as the students who already taken the course are not getting the
benefit of the course as their issues were not solved. In order to benefit the students, those who are currently taking the
course, their feedback should be taken in real time and their issues must be resolved in real time. This is a beneficial
approach as this allows the teacher and students to resolve the teaching and learning issues. Analyzing the student’s
comments and feedbacks by using the sentiment analysis techniques helps to identify student’s positive, neutral or negative
review towards the current teaching methodology adopted by teacher.

INDEX TERMS Sentiment analysis, Data Mining, Educational Data Mining, Student’s Feedback,
teaching methodology, Education, positive, negative, neutral, polarity.

I. INTRODUCTION usually conducted at the end of semester or at the end


Sentiment Analysis in education is developed to of the term [1], [3], [5]. Gathering feedbacks from
address the problems in education sector. Taking students at the end of semester has a disadvantage of
feedback from students to lecturer allows the students not benefitting the students those who are taking the
to guide the teacher into teaching the course in a best course or that have already taken the course.
possible that the can understand the lecture [1], [2], [3].
Sentiment analysis is a field that works on the basis of This paper aims to address the problem, creation of a
to make sense of a textual comment/feedback/opinion. system is proposed that will analyze the student’s
Comments or opinions can be positive or can be feedback in real time and present those feedbacks to
negative[1], [4]. Analyzing student’s given feedback the teacher. In order to benefit the students who are
using the techniques of Sentiment Analysis will be currently taking the course the feedback from students
helpful to identify positive, negative or even more should be taken in real time and their issues must be
refined behaviors of students, they have towards the addressed in real time. This will be help the teacher to
current teaching [1]. The feedbacks provided by benefit students by resolving their issues in a beneficial
students can highlight different issues students are way. Analyzing the feedback provided by students by
facing with the lecture or with the teacher. Most of the using sentiment analysis techniques will help to
times students do not understand the topic teacher is identify student’s positive, negative and even more
trying to explain, they are not getting the topic because refined emotions that students have towards the current
the teaching methodology adopted by teacher is very teaching methodology adopted by teacher.
difficult to understand, by giving feedbacks students
can deliver their issue to the teacher [4]. II. BACKGROUND
According to Chauhan, Agrawal [2] almost all of the
Educational institute in order to improve the quality of educational institutes, most commonly universities
education takes the feedback from students [1]. The showing a lot of interest to improve the quality of
comments of students then analyzed to read the education by keeping check and balance or by
sentiments of the students to enhance and improve the monitoring teacher’s teaching, student’s learning by
teacher’s teaching performance [1]. The main focus of analyzing the feedback provided by students using
sentiment analysis is to determine the feelings of sentiment analysis. To improve the teaching-learning
students given in the form of text. That feeling might process the maximum involvement of students,
be student’s opinion or attitude. The vital role of the teachers and other stakeholders are required to identify
analysis is to classify the polarity of the feedback or different aspects of education. Another class of study
opinion given in the form of textual data as positive, by Ullah [1] which states the efforts and attempts of the
negative or neutral [3]. Feedbacks from students are educational institutes to gather the student’s feedbacks

VOLUME XX, 2017 1


towards the teacher’s teaching in order to enhance the currently studying a course and facing issues to
teaching performance of a teacher or instructor. understanding a lecture by getting the feedback in real
According to another study by Rajput, Haider [3] the time and solving the issues in real-time [7] .
feedback is an essential part of education to improve
the quality of teacher and learning. Students express In this section, paper’s acceptance and rejection criteria
their issues by giving different comments/feedbacks is defined, article selected for study if the following
these comments are analyzed by using sentiment statements are true: -
analysis that whether the comments are positive or 1. The article must be in English language.
negative. 2. Articles must be related to Sentiment Analysis
or student’s feedback analysis.
The feedback system became very useful in the 3. Articles must be published between 2009 and
educational domain, where the teacher delivers a 2020.
lecture and student’s share their opinions towards that 4. Conference paper will be selected if they are
lecture. The opinions are analyzed to address the
very relevant.
difficulties of students by using sentiment analysis,
where sentiment analysis is a computational study of Article rejected for study if the following statements
people’s like and their dislike about an entity or about are true: -
an object also revealing the polarity of the natural 1. If article related to student’s feedback analysis
sentences/word e.g. positive, negative or neutral [1]. but not to sentiment analysis or vice versa.
Also analyzing students’ comments using sentiment 2. Article related to other than educational
analysis approaches can help to classify that the
domain.
students’ feelings are positive or negative. Students’
3. Article without any methodology evidence.
can mention or highlight the issues they faced during
lecture, through their feedbacks they can indicate their
issues to the lecturer [4]. In the domain of education III. METHODOLOGY
sentiment analysis is used to extract the hidden In this study, Long Short-Term Memory model
knowledge from students through a feedback [2]. The (LSTM) model is implemented for the sentiment
evaluation of teacher and class by students’ now is analysis and text classification. An algorithm is
considered a usual practice in the educational designed for this purpose, explaining the architecture
institutes. This process helps to provide quality of the system shown in Figure 1. System is divided into
education and helps the instructor to distill their five phases 1) First phases is Data preprocessing. 2)
teaching methodology. The sentiment analysis Second is word embedding, 3) LSTM model for the
techniques are used to identify the textual contents that hypothesis testing for the prediction accuracy, 4) Dense
either they are positive or negative provided by layer in order to increase model complexity, 5)
students as feedback [3]. Sentiment analysis helping Softmax function for multi-class classification
the students’ those who have issues to forward their problems.
concerns to the teacher so that he can improve his/her
teaching methodology as a result students’ get quality Data Preprocessing
education and understand the lecture clearly [6].

In the light of above mentioned studies, it is stated that Embedding Layer


the educational institutes used to get the feedback at
the end of semester or year [1], [3], [5]. Taking the
.4 .6 .5 .4 .4
feedback at the end of semester has a disadvantage that .2 .2
.2 .3 .3
it is not benefitting the students those who are taking .1 .1 .3 .1 .1
the course or that have already taken the course. In
order to benefit the students who are currently taking
the course the feedback from students should be taken
in real time and their issues must be addressed in real .9 .6 .5 .4 .5
time. This will be help the teacher to benefit students .2 .3 .3 .2 .2
.1 .1 .3 .1 .1
by resolving their issues in a beneficial way. Analyzing
the feedback provided by students by using sentiment
analysis techniques will help to identify student’s
positive, negative and even more refined emotions that Dense
students have towards the current teaching
methodology adopted by teacher. In brief this study is
expected to contribute to benefit the students who are Softmax

VOLUME XX, 2017 1


Through stop words we are able to remove
non-relevant information by categorizing the article
.8 .7 .1 …………... .9 words into a meaningful context. As a result, the
Figure 1 Sentiment Analysis Model Structure. model accuracy increases through the polarity
assigned to words.
A. Data Source
Nowadays, internet has become one of the major C. Word Embedding
source for people to express their feelings. Users Word Embedding is a type of word
are now more willing to share and express their representation, allowing words having similar
opinions or feedbacks online [8]. There is an meaning to have a similar representation. In a
advantage as more and more opinions can be document, the word embedding is the most
extracted from a wider range of source. In data significant representation of words. The main
mining, social networks have been used for years focus is in maintaining the relationship of words,
[9]. This gave so many good advantages in using capture the context of word in a document in
social media such as twitter as twitter is up-to-
order to identify the semantic and syntactic
date and shares information about current news
similarity. In our proposed model the word
and events happening all over the globe [10]. In this
project the data has to be in real time and for this process is a pre-trained word vector and it is
purpose twitter will be used. The tweets of the given as an input to the Long Short-Term
users are collected from the twitter on the basis Memory (LSTM) network. First of all, these
of that input specified by user is in the form of vectors are given to Word2Vec model of Google
Hashtags [10]. The process of tweet classification and in the second step they are pre-trained with
is initiated first by the collections of the tweets. It is the given model on the updated datasets of
achievable to collect the data of twitter by Google news of 100 billion. A 300-dimentioanal
using a twitter API. RAuth library is used in vectors for 3 million words, phases are produced
performing authentication by giving in keys. A by Word2vec model. Also supports the structure
consumer key, access token, consumer secret of bag of words.
and access token secret for twitter application and
perform Handshake protocol [11]. After all this, a D. Long Short Term Memory (LSTM)
certificate downloaded and to access the tweets a A type of recurrent neural network which is
PIN is generated for the application. capable of learning order dependence in a
sequence prediction problem is call Long Short-
B. Data Preprocessing Term Memory (LSTM) network. The
Once the data is obtained from the source, the data representation of a sentence is done by LSTM
has to be pre-processed in order to extract model in a sequential manner. Word vector works
useful information. This can be done by removing as an input for one stage that fed the LSTM layer
the spelling mistakes, grammar mistakes [12]. and in order to compute the next hidden state,
In this research the following steps for the pre- the previous hidden state is fed to LSTM. The
processing are considered. The process of fundamental advantages for the
filtration, in this process the punctuations, implementation of LSTM for sentence vector is
numbers and other special characters or that to out the fixed length sentence vector for
symbols are removed because they are not giving any random variable length sentences. It also
some useful information, but on the other stores the order of word and it has no
hand creates uncertainty and doubts which can dependence on other linguistic features in order to
mislead from the context. The second process, compute semantics. The predictions are in
is the process of tokenization which is used RNN, sequentially, which assign a memory to the
word categorization from the sentence, network. The results got from the previous
breaking the sentence into words. The case predictions can help to improve the future
convention process which is used after the filtration predictions. LSTM provides an additional feature
and tokenization process, the tokenized words are to RNN which gives RNN a fine-grained control
mutate into lower case. The articles in the over the memory. This feature is responsible
sentences are referred as stop words. After the to control that how much the current input
process of filtration, the tokenization process and concerns in the creation of new memory, and
case conversion process are two processes which sections of memory plays important role in
which plays vital role in the removal of stop output generation. The performance of the
words. These are used for the word connection model is improved by word2vec in the absence
and to make understanding for sentences. of large supervised training set.

VOLUME XX, 2017 1



The following equation shows the flow of LSTM
model where sigmoid logistic function is used
and there are some basic gates, input gate controls
how much new inputs are added into Forget
model, forget gate shows the amount of old data
sent by previous hidden state and the output gate
is used to determine the amount of influence of
current node to the external network.
Following are the equations involved in order to
compute the values of LSTM:

Input Activations REFERENCES AND FOOTNOTES


i t =σ ¿ + W (HI ) (cI) (I ) A. REFERENCES
H t −1 + W C t−1 +¿ b )
1. Ullah, M.A. Sentiment analysis of students
feedback: A study towards optimal tools. in
Input Gate
2016 Internation
f t=σ ¿ + W (Hf ) H t −1 + W (cf ) C t−1 +¿ b(f ))
Workshop on Computational Intelligence (IWCI).
2016. IEEE.
Forget Gate
C t=¿ f t ⨂ C t−1 +¿ i t ⨂ tanh ⁡¿+ 2. Chauhan, G.S., P. Agrawal, and Y.K. Meena,
( Hc ) (c ) Aspect-based sentiment analysis of students’
W H t +b ¿ feedback to improve teaching–learning
process, in Information and Communication
Output Gate
Technology for Intelligent Systems. 2019,
o t=σ ¿ + W (Ho) H t −1 + W (co) C t−1 +¿ b(o))
Springer. p. 259-266.
3. Rajput, Q., S. Haider, and S. Ghani, Lexicon-
ht =o t ⨂ tanh ⁡(C t ) based sentiment analysis of teachers’
evaluation. Applied Computational
The hyper parameters are, LSTM layer has 196 Intelligence and Soft Computing, 2016. 2016.
nodes, it is the output dimension of word 4. Aung, K.Z. and N.N. Myo. Sentiment analysis
vector. In order to train model varies parameters of students' comment using lexicon based
are used. The dropout rate is 0.2 and softmax approach. in 2017 IEEE/ACIS 16th
is used as activation function. Adam optimization
international conference on computer and
function with batch size 64 is used for model
training, on the dense layer softmax activation information science (ICIS). 2017. IEEE.
function for multi-classification is used. In order 5. Nasim, Z., Q. Rajput, and S. Haider. Sentiment
to avoid overfitting dropout regularization is used. analysis of student feedback using machine
The text feedbacks that are inputs fed to the learning and lexicon based approaches. in
embedding layer, converting each word to 300- 2017 international conference on research
dimensional vector. The word embedding
and innovation in information systems
layer parameters are maximum features,
embedding features, input length and (ICRIIS). 2017. IEEE.
vector is fed to LSTM model. LSTM layer 6. Sivakumar, M. and U.S. Reddy. Aspect based
forwards the output to the dense layer for the sentiment analysis⨂of students opinion using
predictions of output. Categorical cross- machine learning techniques. in 2017
entropy for multiclass sentiment classification is used International Conference on Inventive
as the loss function.
Computing and Informatics (ICICI). 2017.
IEEE.
Output Ot
7. Go, A., R. Bhayani, and L. Huang, Twitter
It
Input sentiment classification using distant
LSTM block supervision. CS224N project report, Stanford,
2009. 1(12): p. 2009.
⨂ Ct tanh
⨂ ht
tanh

VOLUME XX, 2017 Ft 1


8. Zuo, M., et al. Data mining strategies and
techniques of internet education public
sentiment monitoring and analysis system. in
2010 2nd International Conference on Future
Computer and Communication. 2010. IEEE.
9. Groot, R.d., Data mining for tweet sentiment
classification. 2012.
10. Sriram, B., et al. Short text classification in
twitter to improve information filtering. in
Proceedings of the 33rd international ACM
SIGIR conference on Research and
development in information retrieval. 2010.

VOLUME XX, 2017 1

You might also like