A Survey of Sentimental Analysis on Zomato Restaurant

Reviews
Deep Kalbande1, Pushpak Patil2, Sahil Kale3, Siddhant Kasar4, Sanjeevkumar Angadi5, Prasad Dhore6
1,2,3,4
Student, Department of Computer Science and Engineering, Nutan College of Engineering and Research, Talegaon Dabhade, Pune, India
5,6
Assistant Professor, Department of Computer Science and Engineering, Nutan College of Engineering and Research, Talegaon Dabhade, Pune, India
DOI: 10.47750/pnr.2022.13.S07.910

The services and products being provided in today’s time are open to all users to have and post their opinions on various platforms. Users
and service providers can be benefitted if a tool is developed to help them understand the sentiments behind the data presented to them. The
practice of identifying the sentiment or attitude indicated in a document is known as sentiment analysis, commonly referred to as opinion
mining. When choosing products or services, buyers and users make important decisions based on their feelings. Systems for rating,
reviewing, and recommending products can more accurately and effectively respond to client sentiments by applying sentimental analysis.
This can raise client satisfaction and the likelihood that they will use the service or purchase the product again. Based on the comments they
receive; firms and businesses can use sentiment analysis to improve their services or goods. We cover a variety of research that have utilized
machine learning algorithms to analyze restaurant reviews' sentiment, including those that used various datasets but applied the same
methodology. We evaluate and contrast the various methods employed in these publications for sentiment analysis of Zomato reviews.
Logistic Regression, Decision Trees, Naive Bayes, Random Forest, and neural networks like LSTM and GRU are a few of the techniques
we research.

Keywords: Sentimental Analysis, Natural Language Processing, Bi-GRU, Bi-LSTM.

I. INTRODUCTION

Rating and review systems are platforms that allow users to provide feedback and ratings on products or services they have
used. These systems have become an important part of modern e-commerce and online platforms, providing valuable
information for both customers and service providers. Rating and review systems are widely used in a variety of applications,
including online marketplaces, social media, and review websites. One of the most popular applications of rating and review
systems is in the food industry, where customers can rate and review restaurants and their dishes. Platforms like Zomato, Yelp,
and TripAdvisor, are prime examples of how rating and review systems have become a fundamental part of the food industry.
Customers use these platforms to find new restaurants, read reviews from other customers, and make informed decisions about
where to dine. However, it is not only limited to the food industry, but also widely used in other industries such as e-commerce,
hotels, travel, and many more. Customers now rely heavily on rating and review systems as a source of information when
making purchases. Before making a purchase, buyers use these platforms to research products and services. They also use them
to read reviews left by previous customers to get a sense of how they felt about the product or service. Customers can avoid
buying goods or services that might not fulfill their requirements or expectations and make more informed purchases. However,
because these systems produce a lot of data, it is challenging for users to quickly and simply obtain the information they require.
The practice of mechanically identifying the sentiment or attitude indicated in a text is known as sentiment analysis, commonly
referred to as opinion mining. Sentimental analysis can assist in removing the pertinent information from the text and improving
the customers' access to it.

Sentiment analysis can be used to automatically classify reviews as positive, negative, or neutral, which can be helpful
for customers to quickly find the information they need. Additionally, sentiment analysis can be used to extract information
such as the main topics discussed in the review, which can be used to improve search functionality and recommendation
systems. Sentiment analysis can also be used to identify patterns and trends in customer feedback, which can be valuable for
businesses and service providers. For example, if a company receives a high number of negative reviews about a specific
product, it can use this information to improve the product and provide a better experience for customers. Sentiment analysis
can also be used to track changes in customer sentiment over time, which can be used to identify which marketing campaigns

Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022 7549

or promotions are most effective. Moreover, sentiment analysis can also be used to predict customer preferences and behavior,
by analysing their reviews and feedback. This can be used by companies to tailor their products and services to better suit
customer needs, which can lead to increased customer satisfaction and loyalty. It can also be used to segment customers based
on their preferences and behavior, which can be used to improve targeted marketing campaigns. In addition, sentiment analysis
can also be used to identify the reasons why customers are unhappy with a product or service. This can be used to make informed
decisions about what needs to be improved, in order to provide better service and increase customer satisfaction. This can also
help in identifying the key drivers of customer loyalty and retention. Overall, rating and review systems play a critical role in
various industries, and sentimental analysis can help to improve the usability of these systems for customers and service
providers. Sentiment analysis can help customers to quickly and easily find the information they need and can help businesses
to improve their services and products based on customer feedback. It can also be used to gain insights and make data-driven
decisions that can lead to increased customer satisfaction and loyalty, which can ultimately drive business growth.

The remaining sections of this work are structured as follows. An overview of the relevant studies is presented in Section
2. Section 3 presents the discussion and conclusion.

II. LITERATURE SURVEY

The literature survey section will provide an overview of the existing research on the sentimental analysis of rating and
review systems. We will examine various studies that have applied sentimental analysis to rating and review systems, and
analyse the different techniques and methods used.

In [1], the focus is on the use of predictive models for determining restaurant ratings. The use of rating as a primary
parameter for judging a restaurant is discussed, with an emphasis on the importance of rating for both customers and businesses.
The company Zomato is mentioned as a major player in the field of food reviews and its impact on how people browse through
restaurants is highlighted. Different machine learning algorithms such as Support Vector Machine (SVM), Linear Regression,
Decision Tree, Random Forest, XGBoost, ADA Boost are discussed as potential tools for predicting restaurant ratings. It is
stated that these models have the ability to analyse various features of existing restaurants in a city and predict their ratings,
making it an important aspect to consider before making a dining decision. Out of all the algorithms used ADA Boost gave the
best results. It is noted that a lot of research has been conducted on factors that affect sales and the market in the restaurant
industry, with a focus on analysing "dine-scape" factors to improve customer satisfaction levels. The conclusion suggests that
if data for other cities is also collected, accurate predictions could be made for those locations as well. Overall, the research
emphasizes the importance of restaurant ratings and the use of predictive models as a tool for determining these ratings. The
potential impact of these predictions on both customers and businesses is acknowledged, and the potential for further research
in this area is suggested by using data from more cities.

The [2] focuses on the use of machine learning (ML) and deep learning (DL) models, as well as explainable artificial
intelligence (XAI) methods, to predict customer sentiments in the food delivery services (FDS) domain. The paper begins by
highlighting the increased demand for food delivery services during the COVID-19 pandemic, and the importance of customer
reviews as a source of information about a company's performance. The goal of the study was to review the current state of
research in the use of ML, DL and XAI for sentiment analysis in the FDS domain, with a focus on identifying areas for
improvement and future research directions. It notes that while lexicon-based and ML techniques are widely used for predicting
sentiments through customer reviews in the FDS domain, the use of DL techniques is more limited due to the lack of
interpretability and understandability of the decisions made. It also finds that 77% of the models used in the FDS domain are
non-interpretable in nature, which can be a source of concern for organizations. The paper suggests that DL models in other
domains have shown good performance in terms of accuracy, but lack understandability, which can be achieved through the
implementation of XAI techniques such as LIME and SHAP. The survey concludes that future research should focus on
incorporating DL models for sentiment analysis in the FDS domain and implementing XAI techniques to increase the
interpretability and understandability of the models. It also advises using topic categorization techniques. Overall, the paper
emphasizes the usage of XAI to improve the understandability of DL algorithm results making the model accurate and
explainable.

The [3] focuses on the use of sentiment analysis in evaluating customer reviews on social media platforms, specifically
Twitter, in the restaurant industry. It highlights the increasing importance of customer service in the restaurant industry, and
the need for sentiment analysis to understand customer opinions and preferences. It also notes the difficulty of manually

Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022 7550

evaluating a large amount of data in customer reviews on social media platforms. The paper discusses Naive Bayes (NB) as the
data mining algorithm used for sentimental analysis. It describes a system that uses the NB algorithm to perform sentiment
analysis on tweets about restaurants, using a sample of real data collected from Twitter. It segregates the data as negative and
positive by forming matrices for it. The proposed system is implemented using the python programming language. The results
of the sentiment analysis show that customers have mostly positive opinions about restaurants, with precision, recall, accuracy,
and error rate of 68%, 80.07%, 73%, and 27% respectively. The paper also mentions the importance of sentiment analysis in
other fields such as recognizing bloggers, identifying influential users and bloggers, and providing insight into public opinions
on different subjects. Overall, this work highlights the importance of sentiment analysis in understanding customer opinions
and preferences in the restaurant industry, and the potential of using data mining algorithms such as Naive Bayes for this
purpose. The survey also identifies the challenges of evaluating large amounts of data in customer reviews on social media
platforms and suggests areas for future research in this field. One drawback of the model used in this work is that the parameters
are considered independent of each other in NB whereas that’s not the case in real life and it’s predicted that the results can be
improved by using DL and neural networks in the future.

In [4] the increasing trend in online food services in Bangladesh is addressed by the application of NLP techniques.
With the rise of online delivery services, the overall quality of restaurants is now being judged by customer reviews. The
opinions of customers are organized in a structured way to understand the perception of customer reviews and reactions. This
was the main goal of this research. The first step in the research was to collect data from various sources, such as websites. This
data was then used to create a dataset that was deployed with deep learning techniques. In this paper, a combined CNN-LSTM
architecture is used to analyse the sentiment of text for sentences. Previous works in this domain include using a CNN model
for aspect extraction of Bangla reviews, combined machine learning models for predicting reviews, and more. These works
have used rule-based and machine-learning techniques to analyse sentiment, but they have not been able to achieve the best
accuracy. Research shows that deep learning techniques can provide higher accuracy and purification of the data. In this
experiment, a combined CNN-LSTM architecture was introduced. The main contributions of this research are to build a new
technique that will help to get higher accuracy in sentiment analysis, especially in the Bangla language. A dataset of 1000
reviews is collected from sources in Bangladesh to check the performance of the model. These reviews are unstructured and
unlabelled, so the data is first made into a structure and labeled as positive or negative. After some pre-processing, the data is
fed into the applied model. The CNN is used to learn the representation of words, and the LSTM is used to learn higher
representation of words that are specially used for classification tasks. Hyperparameters are also set before training the model
to improve accuracy. And the model gives out satisfactory results. The output could be further improved if word embedding is
done with more dimensions and trying out other algorithms too.

In [5] it has been discussed that in recent years, there has been a growing interest in the field of opinion mining or
sentiment analysis, which aims to classify text written in natural language as positive, negative, or neutral based on the emotions
and sentiments expressed in it. This is particularly relevant in the context of online customer reviews, as they are a valuable
source of information for making decisions, such as choosing a restaurant or hotel. However, manual analysis of such many
reviews is impractical, making automated approaches necessary. Opinion mining can be performed at different levels, including
document level, sentence level, and aspect level. While most previous work has focused on document or sentence-level
sentiment analysis, this paper focuses on aspect-based opinion mining of restaurant reviews, specifically finding important
features or aspects (such as food or service) and their associated sentiment (positive, negative, neutral) to create a sentiment
profile for each restaurant. This paper aims to implement an aspect-based opinion miner for the tourism domain such as
restaurant reviews, which automatically finds important aspects and opinions of a restaurant by analysing reviews, then creates
a sentiment profile of each restaurant, which can be further used to compare and select restaurants at a particular location by a
traveller. The approach suggested in this paper is to use SentiWordNet, two-word phrases, and linguistic rules together and
have an automatic acquisition of aspects. The paper presents the proposed system architecture and describes the sequence of
steps for the implementation of the system. The results can be improved by considering word sense disambiguation which has
been ignored in this work and only explicit aspects were considered.

According to [6] in recent years, the amount of user-generated content on the internet has increased significantly, making
it difficult for businesses to provide personalized services and for users to extract information efficiently. Sentiment analysis of
restaurant reviews is one way to make sense of this unstructured data. The use of Bi-GRU in sentiment analysis has been shown
to be effective, as it guarantees symmetry in the hidden layer weight update, taking context into account in online restaurant
reviews and producing better results with fewer parameters. In this research, a sentiment analysis model for online restaurant
reviews was built by combining Word2vec, Bi-GRU, and the Attention deep learning method. The model was trained and
validated using restaurant reviews from the website Dianping.com. The results of the model showed F1-score greater than 89%,
indicating that the overall performance of the Word2vec+Bi-GRU+Attention sentiment analysis model was better than
commonly used sentiment analysis models. This research applied deep learning methods to sentiment analysis in the context of

7551 Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022

online food ordering platforms to improve the performance of sentiment analysis in the restaurant review domain. A major
drawback of this work is the unequal number of positive and negative reviews which may cause deviations from the desired
results another factor to consider is the ignorance towards the polarity of the reviews which can be addressed in future works
using appropriate clustering and machine learning methods.

In [7], the authors aim to explore the use of online reviews in the restaurant industry to create sustainable marketing
strategies. They use 294,034 Yelp.com reviews to extract the main aspects of the reviews (food/taste, experience, location, and
value) using the Latent Dirichlet Allocation (LDA) method and assign positive or negative sentiment to each aspect. They also
propose a classification algorithm based on Support Vector Machine (SVM) with a Fuzzy Domain Ontology (FDO) algorithm,
which outperforms traditional classification algorithms such as Naive Bayes and SVM ontology in predicting the helpfulness
of online reviews. The utilization of SVM with the FDO algorithm achieved prediction accuracy (79.59%) and precision rate
(81.62%) to predict the restaurant review usefulness of three U.S.-based cities on Yelp. The authors argue that the results of
this study can be used as a sustainable marketing strategy for review website developers to design more sophisticated and
intelligent review systems, and can also help restaurateurs better understand how to meet diverse customers' needs and maintain
sustainable competitive advantages. It states that future efforts should be made keeping the temporal nature of the reviews in
mind and other formats of data that can also be analysed and utilized in a beneficial manner.

In order to investigate the underlying analogies implicating reviewers’ fine dining restaurant experiences, semantic
network analyses across each emotion were carried out by the authors in [8] around Cantonese restaurants. The research uses a
total of 2,118 reviews, and findings indicate that five emotions – joy, sadness, disgust, surprise, and anger – accounted for 72%
of prediction accuracy. The results suggest that restaurants should understand customers’ emotions based on big data analysis
in order to improve their performance. This study contributes to theory development because this study is the first to introduce
cognitive appraisal theory to online review analytics. Most previous studies on analysing online reviews did not adopt theories
but reported results generated using text analysis programs. It helped to conclude that the words having different emotional
meanings were most part of a completely different network. For example, Keywords in the joy network were pertinent to overall
quality, while keywords in the sadness and anger networks pertained to slow service and rude staff. Keywords in the disgust
network were relevant to mediocre food and high price, while keywords in the surprising network represented good view and
wine pairing with the course menu. The results should be that some aspects generated only positive emotions while some only
generated negative emotions. The paper also suggested ensuring the emotion model is being used as a reference as the
definitions of the emotions may vary from model to model. Plutchik’s circumplex emotion model was used for this study. The
results need to be crosschecked as its one of the first attempts in this area and results generated in this work maybe be biased
because of the nature of the location from where the dataset was taken.

The author of [9] aimed to develop an application that would help companies and users to get more valuable information
out of customer feedback systems. The authors studied the relationship between customer reviews and the company’s overall
performance and made further efforts in that direction. In this paper, the researchers have used a tagged corpus to train a
sequence classifier such as a Conditional Random Field (CRF). For aspect-level sentiment analysis, the important first step is
to identify the aspects and their associated entities present in customer reviews. Aspects can be either explicit or implicit, where
the identification of the latter is more difficult. For restaurant reviews, this difficulty is escalated due to the vast number of
entities and aspects present in reviews. A correlation between the two aspects was concluded and the importance of the
sentiments of the reviews was brought to light. Aspect-based sentimental analysis was used to get the desired results. The app
developed in this paper can help companies to receive a good idea about the customer sentiment about their product and/or
service, and to make more detailed and productive strategies for product creation and marketing. The accuracy rates of the
algorithms used in this paper were also investigated, and the results show that an app is a useful tool for businesses to improve
their customer service and product quality.

A framework of the cycle for writing and analysing restaurant reviews is proposed in [10]. The method is designed to
extract evaluation expressions specific to dishes from the descriptions around the names of dishes in restaurant reviews. It
focuses on two types of expressions that frequently occurred with the name of the dish: common expressions for flavors such
as sizzle words, and words related to dishes such as cooking methods. Also, the expressions evaluating a dish are acquired by
using the dependency analysis and the word importance. In our consideration, the evaluation expression specific to the dish
seemed to capture the characteristics of the dish better than just using sizzle words. The framework formed in this work is very
basic in nature and just makes use of the fundamentals of natural language processing the future developments of the model
mention word embedding techniques and a multi-lingual system which would help in getting better results for language-specific
words.

Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022 7552

 In [11] a new sentiment analysis approach for restaurant reviews aims to address the limitations of existing senti-
lexicons and improve the accuracy of sentiment classification. The research focuses on the problem of a gap in accuracy between
positive and negative sentiment classification using traditional algorithms such as Naive Bayes and SVM. To solve this problem,
the authors propose a new senti-lexicon that is specifically tailored to restaurant reviews and an improved Naive Bayes
algorithm that takes into account both unigrams and bigram patterns in the lexicon. It first describes the research background
and the challenges of sentiment analysis in the restaurant review domain, including the mismatch between the star score and
the review text, and the limitations of existing lexicons for this specific domain. It includes the construction of a new senti-
lexicon that includes sentiment words associated with restaurant reviews, and the development of an improved Naive Bayes
algorithm that considers both unigrams and bigrams patterns in the lexicon. The paper compares the classification of
performance and accuracy of the original Naive Bayes and SVM algorithms with their improved version. The results show that
the proposed approach narrows the gap between positive and negative classification accuracy, and the difference between the
positive classification accuracy and the negative classification accuracy was narrowed by 3.6% compared to when the original
Naïve Bayes was used, and there was a difference of 28.5% compared to when SVM was used. Overall, this paper presents a
novel approach to sentiment analysis of restaurant reviews that address the limitations of existing senti-lexicons and improves
the accuracy of sentiment classification by considering both unigrams and bigram patterns in the lexicon. The model has the
potential to provide more accurate and reliable sentiment analysis of restaurant reviews and help businesses to understand
customer sentiment and improve their services.

In [12] a new approach for movie recommendation in a mobile environment that aims to improve the accuracy and
timeliness of mobile movie recommender systems is shown. The proposed method uses a combination of a hybrid
recommendation model and sentiment analysis on the Spark platform. The hybrid recommendation model is used to generate a
preliminary recommendation list, which is then optimized using sentiment analysis. This approach takes into account the user's
preferences, reviews, and emotions to help them find suitable movies conveniently. It starts with highlighting the importance
of movie recommendation in the mobile environment, which is becoming increasingly critical as people's daily lives become
more dependent on mobile services. The authors point out that movie recommendation is a comprehensive and complicated
task that involves various tastes of users, various genres of movies, and so forth. Therefore, many techniques for
recommendation have been proposed over the years to solve the problems. It mentions the traditional methods like content-
based recommender systems, collaborative filtering recommender systems, and hybrid recommender systems, each with its
own advantages and limitations. It then goes on to describe the proposed method, which is a hybrid recommendation model
with sentiment analysis, implemented on the Spark platform. The proposed method is able to improve the accuracy and
timeliness of mobile movie recommender systems by combining the advantages of collaborative filtering and sentiment
analysis. Moreover, the sentiment in the reviews is an important aspect in evaluating a movie, as it can give a better
understanding of the user's preferences and feelings about the movie. It presents the results of the evaluation of the proposed
method, which was compared to two other summary methods, the Refresh and Gensim systems, based on criteria such as
informativeness, clarity, helpfulness, and likes. The results showed that the proposed method outperforms the other two
methods, proving the effectiveness of the hybrid recommendation model with sentiment analysis. Finally, the paper concludes
that the proposed method can provide sufficient and clear information about the target movie, and is helpful for users to
understand the actual situation of the movie. It also suggests that future work will focus on eliminating individual characteristics
hidden in the text description from users and making use of the word data of sentiments to find the sentiments rather than
relying on ratings. The proposed method is also well received by users as it is able to provide a convenient and fast way for
them to obtain useful movie suggestions. Overall, the research paper presents a new approach for movie recommendation in a
mobile environment that uses a hybrid recommendation model and sentiment analysis to improve the accuracy and timeliness
of mobile movie recommender systems, and it shows that it can be an effective tool for movie recommendation on mobile
services.

In [13] the growing importance of online food delivery (OFD) during the COVID-19 crisis and the importance of
customer reviews as a source of data for OFD companies is shown. The paper suggests that OFD companies place significant
importance on collecting complaints from customer feedback and using data effectively to identify areas for improvement to
increase customer satisfaction. The paper suggests that online reviews remain important during the COVID-19 pandemic as
they help customers make safe food decisions. The paper proposes a Natural Language Processing (NLP) based approach,
specifically Sentiment Analysis, as a method for OFD companies to evaluate customer satisfaction and complaints. Sentiment
Analysis is a technique that can be used to categorize emotions as positive, negative, or neutral. The study applied Unigram,
Bigram, and Trigram to perform the sentiment analysis and found that the main issues with OFD companies are primarily
related to food delivery issues and that both organizations generally experience the same issues. The study suggests that the
proposed method can be used as a guide for OFD companies to evaluate customer satisfaction and complaints and develop
marketing strategies to acquire new customers and increase their market share. The paper concludes that online reviews are an
important source of data for OFD companies during the COVID-19 pandemic and that Sentiment Analysis is an effective
method for evaluating customer feedback and identifying areas for improvement.

7553 Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022

 In [14] a study that focuses on analysing restaurant reviews to provide useful information that is not captured by
traditional ratings is shown. The study uses two different datasets of restaurant reviews and employs machine learning
algorithms such as Naive Bayes and Logistic Regression to classify the reviews and perform sentiment analysis. The study also
utilizes the gensim library for summarization and uses visualization techniques to display the results. The study found that the
Multinomial NB model performed better than other algorithms in terms of accuracy, precision, recall, and F1 score. The study
also found that taking online orders and table bookings increases the number of customers, and suggests that restaurants should
collaborate with delivery sites like Zomato and Swiggy to grow their business. Additionally, the study suggests that restaurants
should offer quick bites for areas where there are many MNCs and that the average cost for a person should not be high. In the
future, a real-time application can be developed to help people analyse reviews more effectively to grow their businesses. Neural
network architectures can be used to make text summarization more accurate and readable. The ability to gain insights from
reviews can open up a new world in the field of analytics and how data is used in businesses. There is potential for future
research in this area, and the study can be applied to various industries such as transportation, hospitality, healthcare, and
education. The study concludes that analysing restaurant reviews in an effective manner is crucial for restaurants to be
successful and prosperous and that there is potential for future work to develop an efficient analysis system utilizing the potential
of reviews.

In [15] an abstractive multi-text summary method for generating restaurant reviews is shown. The proposed method
uses a combination of natural language processing techniques, including the TextRank algorithm, to automatically extract the
most informative and representative sentences from the reviews, based on predefined topics and sentiments. The method aims
to provide users with a comprehensive and easily digestible summary of the restaurant, including basic information, key positive
and negative aspects, and an overall sentiment ratio. The researchers used four predefined topics to classify the information,
which is chosen based on the common choices that most users look for on review websites. The method was evaluated and
compared to two other summary methods the Refresh and Gensim systems, using various criteria, such as informativeness,
clarity, helpfulness, and likes, and it was found to be superior in all aspects. The study concludes that the proposed method can
be an effective tool for quickly understanding the overall sentiment and key aspects of a restaurant, and has a potential for future
expansion to other areas and aspects.

A machine learning-based approach for sentiment analysis of restaurant reviews is proposed in [16]. Sentiment analysis
determines the emotional tone, attitude, and sentiment expressed in each text. A dataset of restaurant reviews is used and split
into two parts: 70% for training the model and 30% for testing the performance. They used the Support Vector Machine (SVM)
algorithm to train the model and achieved an accuracy of 79%, a precision score of 0.8, and a recall score of 0.73. the model
can be further extracted and hosted on the internet for sentiment analysis. It also shows the different sentiment analysis
approaches, including lexicon-based, machine learning-based, and hybrid approaches. A lexicon-based approach uses
dictionaries of polarity words to determine the overall polarity of a sentence. A machine learning-based approach uses
supervised classifiers that are trained on previously labeled information, and a hybrid approach combines both lexicon-based
and machine learning-based approaches. The conclusion of the paper is that the machine learning-based approach is more
reliable and robust than a lexicon-based approach. They also mention that the model is domain-specific and cannot be used for
sentiment analysis of reviews from different domains. However, the model can be further improved by using other techniques
such as data augmentation or over-sampling/under-sampling methods to balance the

In [17] the underlying factors of customer value in a restaurant setting by applying machine learning-based natural
language processing techniques to the analysis of a vast amount of online customer reviews explored. It identifies 14 factors
that reflect a holistic view of previous research on customer value. The findings suggest that the comprehensive approach
incorporating cognitive and affective aspects into the experiential perspective of value gains deeper insights into the nature of
customer value. Moreover, the study uniquely finds new factors that prior research has rarely investigated. It proposes a
methodological framework that enables researchers to test the quantitative measure of customer perceptions derived from
unstructured online reviews. It finds that service providers need to fully understand specific factors that constitute customer
value to receive customer preference. The 14 specific factors identified in the study contain customers’ actual experiences,
which service providers can use to improve their service operations. For example, service providers may be able to develop
performance indices based on those identified factors. Additionally, the study also identifies many negative factors of customer
value and suggests that practitioners need to not only improve the positive factors but also mitigate the negative factors, given
that they are related to customer satisfaction. It also suggests that hospitality and social media firms need to adopt a holistic
view of customer value. When developing a service design or marketing promotion, managers often focus on the benefits that
customers can cognitively perceive. However, findings in this study suggest that they should not ignore either the cognitive or
affective factors of customer value. Social media firms providing online customer reviews, they will be able to restructure
customer reviews and rating systems to provide critical factors broadly and develop a recommendation system that matches
customer preference with the factors that particular restaurants satisfy. It is limited to the food service industry and does not

Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022 7554

specify the type of restaurant to find a difference. However, the study has identified some new value factors that are likely to
be worth future in-depth empirical research and qualitative and quantitative studies should be conducted to further understand
the value perception by validating the findings.

The [18] delves into the recent studies and research done in the field of sentiment mining, specifically in the context of
automating the sentiment analysis process. It would show the various methods that have been tested and propose a new hybrid
classification method based on coupling classification methods using an arcing classifier. It also covers the different algorithms
used in the proposed method, such as Naïve Bayes (NB), Support Vector Machine (SVM), and Genetic Algorithm (GA), and
their performance in terms of accuracy. Additionally, it also shows the effectiveness of ensemble techniques for sentiment
classification and the feasibility and benefits of the proposed approach using restaurant reviews as a case study. This paper also
includes the studies that have been done on extracting classification information from reviews and using them to understand
the context behind a rating and the importance of this in helping users make more informed decisions. The accuracy of the
model using algorithms like Naïve Bayes, Support Vector Machine, genetic algorithm, and Proposed Hybrid Methodology is
85%, 85.20%, 85.30%, and 92.44% respectively. Comparison between the individual classifier and the hybrid classifier: the
hybrid classifier shows a significant improvement over the single classifiers.

In [19] the use of sentiment analysis in natural language processing to analyse reviews of restaurants on the Zomato
application is shown. The researchers pre-processed the data by making all words lowercase, tokenization, removing numbers
and punctuation, removing stop words, and lemmatization. They then used the term frequency-inverse document frequency
(TF-IDF) to create word vectors. 150,000 reviews were used in the study. The data was then split into positive, negative, and
neutral sentiments based on the rating of the review. The model used was a random forest classifier and the metrics used to
evaluate the model were precision, recall, and accuracy. The accuracy of the model was found to be 92.43667%. The precision
of positive, negative, and neutral sentiments was 92%, 93%, and 96% respectively. The recall of positive, negative, and neutral
sentiments was 99%, 89%, and 73% respectively. The average precision and recall were 93% and 87%. The 10 words that had
the most impact on the results were: "bad", "good", "average", "best", "place", "love", "order", "food", "try", and "nice". The
study concludes that the model is effective in predicting the sentiment of restaurant reviews, but suggests that further research
can be done to improve the results, such as using imbalanced dataset algorithms and using the word data of sentiments to find
the sentiments rather than relying on ratings.

III. CONCLUSION

This survey aims to explore the current state-of-the-art techniques and methods used for improving the rating system in
Zomato, a popular online food delivery and restaurant review platform, using word processing techniques such as word2vec
and machine learning algorithms like bi-LSTM. The online food delivery and restaurant review industry have been rapidly
growing in recent years, and it has become crucial for businesses to understand customer sentiment towards their restaurants.
Sentiment analysis is a technique that uses natural language processing and machine learning to understand customer emotions
and opinions, and it can be used to make data-driven decisions and improve the overall customer experience. The survey will
provide an overview of existing research, discuss the potential benefits and limitations of these techniques, and present a
systematic review of studies that have applied word2vec and bi-LSTM to improve the rating system in Zomato. The main
objective is to provide a comprehensive understanding of current research trends and advancements in this field and to serve as
a starting point for future research. Despite the success of the bi-GRU approach, which is currently the most popular technique
used, there is still room for improvement, and it is necessary to explore new techniques and methods for sentiment analysis
with the increasing amount of data and the need for more accurate analysis. This survey will help researchers to understand the
current trends and advancements in the field and to make better data-driven decisions.

REFERENCES
1. Atharva Kulkarni, Divya Bhandari, Sachin Bhoite (2019) “Zomato Restaurants Rating Prediction using Machine Learning Algorithms”.
2. Anirban Adak, Biswajeet Pradhan, Nagesh Shukla (2019), “Sentiment Analysis of Zomato Restaurant Reviews of Food Delivery Services Using Deep
Learning and Explainable Artificial Intelligence: Systematic Review”
3. Murtadha M. Hamad, Mohanad Ahmed Salih and Refed Adnan Jaleel (2021), “Sentiment Analysis of Restaurant Reviews in Social Media using Naïve
Bayes”.
4. Naimul Hossain, Md. Rafiuzzaman Bhuiyan, Zerin Nasrin Tumpa (2020), “Sentiment Analysis of Restaurant Reviews using Combined CNN-LSTM”.
5. Chinsha T C, Shibily Joseph, (2014),” Aspect based Opinion Mining from Restaurant Reviews”.

7555 Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022

6. Liangqiang Li, Liang Yang and Yuyang Zeng 2021, “Improving Sentiment Classification of Restaurant Reviews with Attention-Based Bi-GRU Neural
Network”
7. Yi Luo and Xiaowei Xu 2019, “Predicting the Helpfulness of Online Restaurant Reviews Using Different Machine Learning Algorithms”
8. Munhyang Oh & Seongseop Kim 2021, “Role of Emotions in Fine Dining Restaurant Online Reviews: The Applications of Semantic Network Analysis
and a Machine Learning Algorithm”
9. Aaditya Prakash Pillai, Dr. S. P. Chauhan 2022, “Sentiment analysis of customer reviews-using ai/ml approach”
10. Ryoya Suzuki, Ryosuke Yamanishi, Yoko Nishihara 2019, “Framework for Analysis and Feedback of Evaluation Expressions in Restaurant Reviews”
11. Hanhoon Kang, Seong Joon Yoo,Dongil Han 2012, “Senti-lexicon and improved Naïve Bayes algorithms for sentiment analysis of restaurant reviews”
12. Yibo Wang, Mingming Wang, and Wei Xu 2017, “A Sentiment-Enhanced Hybrid Recommender System for Movie Recommendation: A Big Data
Analytics Framework”
13. KevserSahinbas, Arda AVCI 2022, “Sentiment analysis of restaurant review in online food delivery industry”
14. Dhiraj Kumar, Gopesh ,Avinash Choubey, Pratibha Singh 2020 , “Restaurant Review Classification and Analysis”
15. Yenliangchen, chialingchang, and jeryeu gan1 2021, “A Template Approach for Summarizing Restaurant Reviews”
16. Vishal Lokam, Vaishnavi Shinde, Abhishek Raikar, Vivek Kate, Gaurav Jumnake 2021, “Restaurant and Cuisine Review Sentiment Analysis using SVM”.
17. Wooseok Kwon, Minwoo Lee, Ki-Joon Back 2020, “Exploring the underlying factors of customer value in restaurants: Amachine learning approach”.
18. M. Govindarajan 2014, “Sentiment analysis of restaurant reviews using hybrid classification method”.
19. Bern Jonathan, Jay Idoan Sihotang, Stanley Martin 2019, “Sentiment Analysis of Customer Reviews in Zomato Bangalore Restaurants Using Random
Forest Classifier”

Journal of Pharmaceutical Negative Results ¦ Volume 13 ¦ Special Issue 07 ¦ 2022 7556