VW,QWHUQDWLRQDO$UDE&RQIHUHQFHRQ,QIRUPDWLRQ7HFKQRORJ\ $&,7

Sentiment Analysis and Classification of

Restaurant Reviews using Machine Learning
Kanwal Zahoor, Narmeen Zakaria Bawany, Soomaiya Hamid
Center for Computing Research, Department of Computer Science and Software Engineering
Jinnah University for Women

Abstract— In the last few years, use of social networking Customer sentiment analysis [6] is also referred to as
sites has increased tremendously. People use social media emotion AI or opinion mining. It shows the polarity of
platforms to share their views on almost all subjects. These customer reviews [7] i.e. positive or negative. It can also
views are in various forms like, blogs, tweets, Facebook posts, predict ratings [8] for restaurant services based on various
online discussion boards, Instagram posts, etc. Sentiment attributes.
analysis deals with the process of computationally defining and
classifying the views expressed in the comment, post or One of the basic technique used for sentiment analysis
document. Typically, the aim of sentiment analysis is to find is Natural Language Processing (NLP) along with different
out the customer’s attitude towards a product or service. machine learning algorithms. NLP refers to the AI method
Customers’ feedback is vital for businesses, and social media of communicating with an intelligent system using a natural
being a powerful platform, can be used to improve and language such as English [9]. It includes different methods
enhance business opportunities if the feedback on social media for data preprocessing such as; lexical analysis, syntactic
can be analyzed timely. Therefore, the focus of this paper is to analysis, semantic analysis, disclosure integration, and
analyze the customer reviews about various restaurants across Pragmatic analysis. [10].
Karachi – one of the biggest cities of Pakistan. For this
research, customer reviews are collected from a very popular Various studies [11-13] have been conducted for
Facebook community- the SWOT'S guide to Karachi’s sentiment analysis of restaurant customer reviews using
restaurants. The contribution of this research is twofold. First, NLP and machine learning techniques, however limited
it performs sentiment analysis and classifies each comment as research has been performed that targets Pakistani
positive, negative. Second, by using text categorization restaurants. Therefore this research presents a framework
techniques, comments are automatically classified according for analyzing sentiments from customer reviews and then
to feedback about food taste, ambiance, service, and value for predicting the impact of a restaurant on their customers, that
money. A manually annotated dataset of around 4000 records is; good or bad.
was used for training and testing. Several algorithms were
used for classification, including Naive Bayes Classifier, For this research, we have worked on structured data of
Logistic Regression, Support Vector Machine (SVM), and a well-known social eatery community of Pakistan the
Random Forest. The performance comparison of these “SWOT’s guide to Karachi’s restaurants Cafes Dhabas
algorithms is presented. The best results, that is 95% accuracy, HBFE & Takeouts.” This community has more than
were achieved by using a random forest algorithm. 400,000 active members till date. Mostly the members of
this community share their experience about restaurants and
Keywords— Sentiment Analysis, Category-Classification, their services. We have created an annotated dataset of
Naïve Bayes Classifier, Logistic Regression, Support Vector approx. 4,000 records, to analyze the reviews for achieving
Machine, Random Forest, Natural Language Processing (NLP), sentiment analysis. These records are classified into four
Restaurant Reviews Classification, and Machine Learning
categories, food taste, services, ambiance, and value for
I. INTRODUCTION money.
From the past few decades, the massive use of the The aim of this research is to analyze the reviews and
internet has led to an increase in online activities such as; classify them with respect to category. For this purpose,
social media communication, e-commerce, blogging, NLP techniques were used for data cleaning which are
surfing, etc [1]. This usage has also created a huge impact lexical analysis, syntactic analysis, and semantic analysis.
on different business. Recently, a new trend has been set up After that four classification algorithms, Naïve Bayesian
in which customers prefer to check reviews of a particular Classifier (NBC), Logistic Regression (LR), Support Vector
product before buying it. As in today’s excessively, socially Machine (SVM), and Random Forest, were applied. To
connected world, people trust more on consumer's original evaluate the performance of the model various metrics such
reviews rather than believing in fancy promotional as Precision, Recall, F-score, and Accuracy were used.
advertisements [2]. This trend has been very helpful for the The paper is structured as follows: Section 2 discusses
restaurant’s customers as it becomes easier to select a good the brief literature review of the previous work in the related
restaurant for specific cuisine. [3]. It also help customers to classification problems. Section 3 describes the
identify the best dining place, in terms of taste, ambiance, methodology. Section 4 presents the results while Section 5
and customer service. [4] concludes this research study.
Consequently, it enforces restaurant owners to collect
and monitor customers’ feedback on social media platforms II. RELATED WORK
[5]. Sentiment analysis of customers’ review also helps in A brief survey of the previously performed sentiment
improving the overall customers experience by analysis is presented in this section. Rajasekaran et al. [14]
incorporating the customer suggestions. performed sentiment analysis using Naïve Bayes and SVM

‹,(((
Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.
on restaurant reviews and achieved 72.2% accuracy.
Krishna et al. [15] also analyzed restaurant reviews by using
SVM with 94.56 % accuracy.

TABLE I. RELATED RESEARCH WORK TO PERFORM SENTIMENT ANALYSIS

Sentiment Category
Reference Dataset Algorithms or Models Accuracy
Analysis Classification

[11] Restaurant Reviews ᅛ 8 Naïve Bayes, SVM 72.2%

[12] Restaurant Reviews ᅛ 8 SVM 94.56%

[13] App Reviews ᅛ 8 SentiWordNet, SVM 89%

Tour and Travel NLP technique with CHAID

[14]
Reviews ᅛ 8 decision tree
66.86%

Chinese Restaurant
[15]
Reviews ᅛ 8 Naïve Bayes algorithm, SVM 93.17%

Maximum Entropy, Conditional 80.07%

[16] Trip Advisor Reviews ᅛ 8 Random Field, SVM 75.22%

72.06%
[17] Trip Advisor Reviews ᅛ 8 Naïve Bayes, TextBlob
69.12%
English French and
[18] Greek Restaurant ᅛ 8 SVM 91%
Reviews
Spanish and English SVM, Naïve Bayes, Multinomial
[19]
Restaurant Reviews ᅛ 8 Naïve Bayes
74%

[20] Trip Advisor Reviews ᅛ 8 Decision Tree-J48 45.6%

Naïve Bayesian Classifier, Logistic

Our SWOT’s Karachi
ᅛ ᅛ Regression, SVM, and Random 95%
Research Restaurant Reviews
Forest algorithms

Fikri et al. [16] performed a rule-based sentiment et al. [23] performed sentiment analysis on the data having
analysis with the help of SentiWordNet and SVM with 89% people's perceptions towards the restaurants’ brand image.
accuracy. The customers’ app review data was acquired For this analysis Decision Tree-J48 method was used and
from Play Store and App Store. Guerreiro et al. [17] used achieved 45.6% accurate results.
tours and travel data reviews to perform sentiment analysis
In this research study, we have performed sentiment
using NLP technique with the Chi-Squared Automatic
analysis and category classification on Karachi food and
Interaction Detection (CHAID) decision tree and binary
restaurant reviews. For this purpose, we have used NLP
logistic regression algorithm to predict the sentiments of
techniques and supervised learning algorithms.
customers. In this study highest accuracy was recorded at
66.86% using binary logistic regression. Zhang et al. [18] III. METHODOLOGY
found better accuracy from the Naïve Bayes algorithm than
The analysis conducted in this research comprises two
SVM which is 93.17%. Sentiment analysis was performed
phases. In the first phase sentiment analysis is performed
on Chinese restaurant reviews. Akhtar et al. [19] presented
and in the second phase reviews are classified according to
aspect-based sentiment analysis by using Particle Swarm
the category. For NLP, three techniques has been employed,
Optimization (PSO) with the learning algorithms;
that is lexical analysis, semantic analysis, and syntactic
Maximum Entropy, Conditional Random Field, and SVM.
analysis. Afterward, we have applied four supervised
The accuracy was calculated in two aspects one has 80.07%
machine learning algorithms for sentiment analysis and
while the other is 75.22%. Laksono et al. [20] created their
category classification. Fig. 1 shows the complete
testbed using WEKA having data from TripAdvisor. Naïve
architecture of our research methodology.
Bayes and TextBlob used for sentiment analysis at the rate
of 72.06% and 69.12% accuracy was achieved. Mabrouk et 3.1Dataset
al. [21] used deep learning convolutional neural networks The data used in this research was collected from a
for sentiment analysis. SVM algorithm was used for social media community named as “SWOT’s Guide to
analyzing restaurant reviews and 91% accuracy was KARACHI’s Restaurants Cafes Dhabas HBFE &
achieved. Viloria et al. [22] used NLP for sentiment analysis Takeouts”. The dataset consists of approx. 4000 customer
of restaurant reviews. This research approach includes posts in English language about different restaurants of
SVM, Naïve Bayes, and Multinomial Naïve Bayes. Highest Karachi. Reviews were classified has positive and negative,
accuracy was achieved by 74% using Naïve Bayes. Masrur

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.
based on the nature of the review. Reviews were classified using NLTK component Corpus, which transforms the data
in four categories, that is; food taste, services, value for language into a simplified, independent from words
money, and ambiance. Data count for each category is relationships, and free from grammatical dependencies.
shown in Table. II. After this process, a problem arises in which the words are
scattered and sentences become meaningless. In order to
TABLE II.  DATASET FOR CATEGORY-CLASSIFICATION resolve this issue, we have used
S.No. Label Count “Nltk.WordNetLemmatizer” to combine inflected forms of
1. Food taste 2545
words in a meaningful sentence, so that they can be
2. Service 805
3. Value for Money 447 analyzed as a single instance. For example, the original
4. Ambiance 229 review was [“it was delightfully rich and reasonably
satisfying in taste”] and after the lemmatization, it
3.2Data Preprocessing transforms into [“delightfully rich reasonably satisfying
Data preprocessing is an initial and major step for taste”].
sentiment analysis because customer provides their To finalize the data preprocessing steps we have
feedback or sentiments using natural language syntax. In converted words in root words by eliminating prefix and
order to express their emotions, they use extra letters, suffix. For this purpose we have used the NLTK component,
alphabets, words, and special characters. For classification, named; “Nltk.Stem”. This component works on the Porter
Stemmer algorithm. For example in our dataset, ‘Tasting’
or ‘Tastes’ converts into ‘Taste’. ‘Trying’ or ‘Tried’ convert
into ‘Try’. Because in sentiment analysis there is no need to
focus on grammatical issues we are only gathering true
sentiments via actual word analysis. To support these
components of NLTK we have applied “Term Frequency-
Inverse Document Frequency” (TF-IDF). It is a popular
term-weighting scheme [26] which is used for text mining
to identify the relevancy of the terms in a dataset. After
eliminating all the ambiguities, we prepared bag-of-words
which represents sentences with root words, independent of
any grammatical and punctuation rules.
3.3Split Training and Testing Dataset
Machine learning mathematical models work on
experience and learning. Therefore, there is a need to train
our model by feeding a training dataset. The dataset is split
into two portions, one for training and other for testing.

data must be cleaned from these extras. Therefore we have We used three different distributions of training and
used NLP techniques for sentiment analysis as well as to testing dataset. In the first stage, we have tried an ideal ratio
prepare data for classification. of dataset distribution, that is; 80:20. Then we have also
tested our results at 75:25 and 70:30 ratio.
Fig. 1. Architecture diagram of our research framework 3.4Tools for Data Analysis and Classification

3.2.1 Natural Language Processing

After segregating data, and fixing the proportion for
training and testing dataset we have applied classification
To process and analyze reviews in natural language, models using the training dataset. For classification, we
NLP provides interaction between human language and have applied four supervised learning algorithms, named;
computer programs [24]. In such scenarios, data Naïve Bayes, Support Vector Machine, Logistic
preprocessing is performed using Natural Language Toolkit Regression, and Random Forest.
(NLTK) [25], which provides a practical implementation of
NLP. Logistic Regression [27] is a classification algorithm,
known for its exponential and log-linear functions. It works
In this regard, data cleaning was performed in five steps, with discrete values and maps the function of any real value
to remove punctuation and special characters from the into 0 and 1. For sentiment analysis the hypothesize shows,
reviews used in the data set. In first step, data was tokenized reviews either positive or negative by using the below
by using the NLTK component “NLTK tokenize”. This functions:
component divides English language sentences into distinct
substrings. For example: [delicious taste full of flavor] 1
would become, [“delicious”, “taste”, “full” ”of”, “flavor”] ℎ ( )= —( )
1+
after processing. In the second step, we removed stop-words For category-classification, we have four classes to
from the data. Stop-words are usually connecting words that identify from the training dataset. Therefore we have a
have no sentiments, for e.g.: it, an, the, with, which, etc. For modification in the logistic regression hypothesis. Now
removing stop words we have used three NLP techniques algorithm assumes that there are n values to predict the
Lexical analysis, Syntactic analysis, and Semantic analysis class. Hence the derived equation become:
as shown in Fig. 1. These techniques were implemented

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.
 ()
( )= ( | ,… , )= ( ) ( | )
∑
The Naïve Bayes classifier [28] is based on a conditional
probability model. In sentiment classification two vectors
are involved, so that classifier assumes independence of two
features, and probability is shown as:
P(B|A)P(A)
P(A|B) =
P(B)
Moreover, this model uses more than one feature to predict
the number of features to predict. So the equation becomes
as:
TABLE III. RESULTS OF SENTIMENT ANALYSIS

Algorithm The ratio of training versus testing dataset

80:20 75:25 70:30

Acc. Acc. Acc.
Sentiment P% R% F% P% R% F% P% R% F%
% % %
Positive 92 91 92 92 93 92 90 92 91
Naïve Bayes 91 92 91
Negative 91 92 92 92 92 92 92 90 91

Logistic Positive 93 90 91 93 91 92 93 91 92
91 92 92
Regression Negative 90 92 92 91 93 92 91 93 92
Positive 95 90 92 93 91 92 91 92 91
SVM 92 93 93
Negative 90 95 92 91 93 92 93 91 92

Random Positive 96 94 95 93 91 92 91 93 92
94 95 95
Forest Negative 94 95 95 91 93 92 93 91 92

TABLE IV. RESULTS OF CATEGORY-CLASSIFICATION

The ratio of training versus testing dataset - 75:25

Algorithm Naïve Bayes Logistic Regression SVM Random Forest

Categories P% R% F% Acc.% P% R% F% Acc.% P% R% F% Acc.% P% R% F% Acc.%

Food Taste 82 99 89 93 98 96 84 99 91 95 97 96

Value for
92 51 66 92 91 93 94 61 74 94 91 92
Money
84 86 89 94
Ambiance 92 62 73 93 85 89 94 67 78 94 90 92

Service 93 62 74 91 87 90 95 68 79 94 90 92

*Where: P: Precision, R: Recall, F: F1- Score, Acc.: Accuracy

SVM [29] is known for its fast and dependable Random Forest [30] forms several decision trees and
classification which resolves two-group classification integrate them so that they can predict more accurately.
problems. The classification is conducted for finding hyper- When the tree starts spreading it creates more randomness
plane between two classes’ positive and negative reviews in to the model. Due to the wide diversity of Random Forest
the model, given as: provides better results and prevents from over fitting.
= … | . + | ∑ ∈
=
For category classification, every category have a B i value
for number of hyper plane that is denoted by s. So that this IV. RESULTS AND ANALYSIS
research model get successful in selecting largest B i value.
After cleaning the dataset using NLP sentiment analysis
= … {ℎ | } and category-classification, we applied supervised learning
algorithms, with cross-validation. In sentiment analysis

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.
reviews were classified into positive and negative while V. CONCLUSION
category-classification reviews were classified into four
categories (food taste, value for money, ambiance, and This paper presents a framework for Pakistani restaurant
service). customer reviews sentiment analysis and category
classification of reviews. For this purpose, the dataset has
For sentiment analysis, it is observed from Table III that the been created by collecting users’ posts from social media
75:25 ratio for training versus testing data has given better
group “SWOT” which contains customer reviews regarding
results. Also, the Random Forest achieved the highest
different Pakistani restaurants. The proposed model uses
accuracy of 95% with the highest precision rate (96%),
recall (95%), and F1 score (95%), among all the algorithms. NLP technique for the tokenization of the textual features
In this research confusion matrix generated for Random from the customer reviews. This framework is implemented
Forest classifier obtained 1925 TP (True Positive) and 1941 using four different classification models i.e. LR, NBC,
TN (True Negative), as shown in Table V. Hence, the total SVM and Random Forest. The maximum accuracy was
of 3866 reviews were correctly classified in this model. achieved by Random Forest algorithm that is, above 90% in
sentiment analysis and category classification. We intend to
TABLE V. RANDOM FOREST CONFUSION MATRIX FOR SENTIMENT collect more data from various social networking sites and
ANALYSIS apply deep learning and neural networks for better
Positive Negative performance in future.
Positive 1925 100 REFERENCES
Negative 84 1941
[1] T. Piñeiro-Otero and X. Martínez-Rolán,
“Understanding Digital Marketing—Basics and
We were able to achieve an accuracy of 94% with Actions,” Springer, Cham, 2016, pp. 37–74.
Random Forest algorithm, using 75:25 training-testing [2] T. Y. Lee and E. T. Bradlow, “Automated Marketing
distribution. According to the Table IV, it was observed that
Research Using Online Customer Reviews,” J. Mark.
all four categories were achieved highest metrics with
Random Forest algorithm. However category ‘food taste’ Res., vol. 48, no. 5, pp. 881–894, Oct. 2011.
crossed 95% in all metrics. The highest precision rate [3] B. Yu, J. Zhou, Y. Zhang, and Y. Cao, “Identifying
(95%), recall (97%), and F1 score (97%), among all the Restaurant Features via Sentiment Analysis on Yelp
algorithms was achieved by ‘food taste’. Reviews,” Sep. 2017.
TABLE VI. CONFUSION MATRIX FOR CATEGORY CLASSIFICATION [4] J. S. Beuscart, K. Mellet, and M. Trespeuch,
“Reactivity without legitimacy? Online consumer
Predicted Food Value for Ambianc
True Taste Money e
Services reviews in the restaurant industry,” J. Cult. Econ., vol.
9, no. 5, pp. 458–475, Sep. 2016.
Food Taste 2479 27 5 29
[5] He, Wu, et al. "A novel social media competitive
Value for
69 378 0 0 analytics framework with sentiment
Money
benchmarks." Information & Management 52.7
Ambiance 29 0 198 2
(2015): 801-812.
Service 83 2 5 713
[6] M. V. Mäntylä, D. Graziotin, and M. Kuutila, “The
evolution of sentiment analysis—A review of research
To summarize the performance of this classification topics, venues, and top cited papers,” Computer
model, confusion matrix obtained after applying Random Science Review, vol. 27. Elsevier Ireland Ltd, pp. 16–
Forest algorithm is presented in Table VI. It depicts that 32, 01-Feb-2018.
‘food taste’ category has the highest accuracy, that is; 97% [7] Q. Sun, J. Niu, Z. Yao, and H. Yan, “Exploring
because it correctly identified 2479 instances out of 2549, eWOM in online customer reviews: Sentiment
whereas, 378 out of 447 instances of ‘Value for Money’ (i.e. analysis at a fine-grained level,” Eng. Appl. Artif.
84% accuracy), 198 out of 229 instances of ‘Ambiance’ (i.e. Intell., vol. 81, pp. 68–78, May 2019.
86.49% accuracy), 713 out of 805 instances of ‘Service’ (i.e.
89% accuracy). [8] K. Ravi and V. Ravi, “A survey on opinion mining and
sentiment analysis: Tasks, approaches and
applications,” Knowledge-Based Syst., vol. 89, pp.
14–46, Nov. 2015.
[9] A. Moreno-Ortiz and J. Fernández-Cruz, “Identifying
Polarity in Financial Texts for Sentiment Analysis: A
Corpus-based Approach,” Procedia - Soc. Behav. Sci.,
vol. 198, pp. 330–338, Jul. 2015.
[10] M. Ghiassi and S. Lee, “A domain transferable lexicon
set for Twitter sentiment analysis using a supervised
machine learning approach,” Expert Syst. Appl., vol.

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.
 106, pp. 197–216, Sep. 2018. “Comparative Analysis Between Different Automatic
Learning Environments for Sentiment Analysis,”
[11] Medhat, W., Hassan, A., & Korashy, H. (2014).
Springer, Cham, 2021, pp. 134–141.
Sentiment analysis algorithms and applications: A
survey. Ain Shams engineering journal, 5(4), 1093- [23] M. Adnan, R. Sarno, and K. R. Sungkono, “Sentiment
1113. Analysis of Restaurant Review with Classification
Approach in the Decision Tree-J48 Algorithm,” in
[12] L. U. Memon, N. Z. Bawany, And J. A. Shamsi, “A
Proceedings 2019 International Seminar on
Comparison of Machine Learning Techniques For
Application for Technology of Information and
Android Malware Detection Using Apache Spark,”
Communication: Industry 4.0: Retrospect, Prospect,
Journal of Engineering Science and Technology 14.3
and Challenges, iSemantic 2019, 2019, pp. 121–126.
(2019): 1572-1586.
[24] L. Marquez, L. Marquez, and J. G. Salgado, “Machine
[13] Raut, Vijay B., and D. D. Londhe. "Opinion mining Learning and Natural Language Processing,” 2000.
and summarization of hotel reviews." In 2014
International Conference on Computational [25] E. Loper and S. Bird, “NLTK: The Natural Language
Intelligence and Communication Networks, pp. 556- Toolkit,” May 2002.
559. IEEE, 2014. [26] J. Beel, B. Gipp, S. Langer, and C. Breitinger,
[14] R. Rajasekaran, U. Kanumuri, M. Siddhardha Kumar, “Research-paper recommender systems: a literature
S. Ramasubbareddy, and S. Ashok, “Sentiment survey,” Int. J. Digit. Libr., vol. 17, no. 4, pp. 305–
analysis of restaurant reviews,” in Smart Innovation, 338, Nov. 2016.
Systems and Technologies, 2019, vol. 105, pp. 383– [27] E. Christodoulou, J. Ma, G. S. Collins, E. W.
390. Steyerberg, J. Y. Verbakel, and B. Van Calster, “A
[15] A. Krishna, V. Akhilesh, A. Aich, and C. Hegde, systematic review shows no performance benefit of
“Sentiment Analysis of Restaurant Reviews Using machine learning over logistic regression for clinical
Machine Learning Techniques,” in Lecture Notes in prediction models,” Journal of Clinical Epidemiology,
Electrical Engineering, 2019, vol. 545, pp. 687–696. vol. 110. Elsevier USA, pp. 12–22, 01-Jun-2019.

[16] M. Fikri and R. Sarno, “A comparative study of [28] Venkatesh, K. V. Ranjitha, and B. S. Venkatesh
sentiment analysis using SVM and Senti Word Net,” Prasad, “Optimization Scheme for Text Classification
Indones. J. Electr. Eng. Comput. Sci., vol. 13, no. 3, Using Machine Learning Naïve Bayes Classifier,” in
pp. 902–909, Mar. 2019. Lecture Notes in Electrical Engineering, 2020, vol.
601, pp. 576–586.
[17] J. Guerreiro and P. Rita, “How to predict explicit
recommendations in online reviews using text mining [29] F. Ali et al., “Transportation sentiment analysis using
and sentiment analysis,” J. Hosp. Tour. Manag., vol. word embedding and ontology-based topic
43, pp. 269–272, Jun. 2020. modeling,” Knowledge-Based Syst., vol. 174, pp. 27–
42, Jun. 2019.
[18] Z. Zhang, Q. Ye, Z. Zhang, and Y. Li, “Sentiment
classification of Internet restaurant reviews written in [30] T. M. Untawale and G. Choudhari, “Implementation
Cantonese,” Expert Syst. Appl., vol. 38, no. 6, pp. of sentiment classification of movie reviews by
7674–7682, Jun. 2011. supervised machine learning approaches,” in
Proceedings of the 3rd International Conference on
[19] M. S. Akhtar, D. Gupta, A. Ekbal, and P. Computing Methodologies and Communication,
Bhattacharyya, “Feature selection and ensemble ICCMC 2019, 2019, pp. 1197–1200.
construction: A two-step method for aspect based
sentiment analysis,” Knowledge-Based Syst., vol. 125,
pp. 116–135, Jun. 2017.
[20] R. A. Laksono, K. R. Sungkono, R. Sarno, and C. S.
Wahyuni, “Sentiment analysis of restaurant customer
reviews on tripadvisor using naïve bayes,” in
Proceedings of 2019 International Conference on
Information and Communication Technology and
Systems, ICTS 2019, 2019, pp. 49–54.
[21] L. Medrouk and A. Pappa, “Deep learning model for
sentiment analysis in multi-lingual corpus,” in Lecture
Notes in Computer Science (including subseries
Lecture Notes in Artificial Intelligence and Lecture
Notes in Bioinformatics), 2017, vol. 10634 LNCS, pp.
205–212.
[22] A. Viloria, N. Varela, J. Vargas, and O. B. P. Lezama,

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:53:07 UTC from IEEE Xplore. Restrictions apply.