A SEMINAR REPORT ON

RESTAURANTS REVIEW ANALYSIS

USING

MACHINE LEARNING AND NATURAL LANGUAGE PROCESSING

BY

ADEYEMO OLUWADAMILARE EMMANUEL

COMPUTER SCIENCE HND2

SUBMITTED IN PARTIAL FULFILMENT FOR REQUIREMENTS OF THE

AWARD OF HIGHER NATIONAL DIPLOMA

IN

COMPUTER SCIENCE

1
 CHAPTER ONE

1.0 Introduction

Restaurant review analysis has emerged as a vital area of focus in the modern food service
industry, driven largely by the rapid digitalization of consumer behavior. With the
proliferation of smartphones, internet access, and social media, customers now have
numerous platforms to express their dining experiences publicly. Platforms such as Yelp,
Google Reviews, TripAdvisor, and even social media sites like Instagram and Twitter have
become hotspots for user-generated content about restaurants. These reviews generate vast
repositories of unstructured data, comprising textual feedback, star ratings, images, and even
video content. For restaurant owners and managers, this influx of data presents both a
valuable opportunity and a daunting challenge. The potential to understand customer
expectations, identify operational issues, and enhance service offerings is immense, but
harnessing these insights from unstructured text requires specialized tools and strategies.

In an increasingly saturated and competitive market, public perception can make or break a
restaurant’s reputation. Consumers often consult online reviews before choosing where to
dine, especially when trying a new place. Positive reviews can enhance a restaurant's
visibility, improve its ranking on review platforms, and contribute to word-of-mouth
marketing, which remains one of the most effective promotional methods. Conversely, a few
negative or poorly handled reviews can significantly damage a restaurant’s credibility and
deter potential customers. This is particularly true in urban settings where diners have a wide
array of options and are likely to be swayed by the latest reviews. As such, maintaining a
positive online presence is no longer optional—it is a strategic necessity that directly
influences customer acquisition and retention.

Customer reviews typically cover a wide array of service dimensions, ranging from the
quality and presentation of the food to the professionalism of the staff, cleanliness of the
dining area, ambience, waiting times, pricing structure, and perceived value for money. These
elements collectively form the basis of customer satisfaction and help shape the overall
dining experience. New restaurants are especially vulnerable to initial reviews, as first
impressions tend to establish long-term expectations among patrons. In many cases, a single
influential review—either glowing or critical—can dictate whether a new restaurant garners
momentum or struggles to attract clientele. Established restaurants, on the other hand, must

2
consistently uphold service standards, as any decline in quality can quickly be reflected in
online feedback, leading to a gradual erosion of customer trust and loyalty.

The nature of customer reviews varies widely. Some comments are highly detailed and offer
specific, actionable insights that management can address immediately—examples include
feedback like “the pasta was too salty,” “the waiter was inattentive,” or “the restroom was not
clean.” These pinpointed concerns allow restaurants to make targeted improvements.
However, other reviews may be vague or emotionally charged without offering clear
guidance, such as “I didn’t enjoy my visit” or “something felt off.” Interpreting such
feedback and determining whether it reflects a one-off incident or a recurring issue is a major
challenge for management. Moreover, when reviews are written in casual or colloquial
language, or when customers use sarcasm, emojis, or local slang, extracting the intended
meaning becomes even more complex.

To effectively manage and interpret this overwhelming volume of user feedback, restaurants
are increasingly turning to advanced computational techniques such as Natural Language
Processing (NLP), sentiment analysis, and machine learning. These tools allow businesses to
automate the process of reading and categorizing reviews, identifying recurring themes,
measuring sentiment intensity, and even predicting future customer behavior based on
historical trends. For example, NLP models can scan thousands of reviews to determine that
customers consistently complain about slow lunch service or praise the weekend brunch
menu. Such insights allow management to take data-informed actions to enhance specific
aspects of the service. This research aims to explore how restaurants can leverage such
analytical methods to isolate key drivers of customer satisfaction. The focus will be on
identifying patterns in the textual content of reviews related to five core themes: food quality,
service efficiency, overall ambiance, pricing, and perceived value.

The digital age has significantly transformed how feedback is gathered and utilized in the
restaurant industry. Traditionally, customer feedback was collected through mechanisms such
as paper-based comment cards, direct verbal communication with staff or managers, and
infrequent survey responses. These methods were limited in scope, anonymous, and often not
actionable in real time. In contrast, the shift toward digital platforms has created an
environment where feedback is continuous, immediate, and visible to a global audience. This
transformation offers restaurants an unprecedented level of insight into customer preferences
and behavior. However, it also introduces new complexities—restaurants must now process

3
and interpret massive volumes of information in near real-time, something that is simply not
feasible through manual effort alone. Therefore, the integration of automated data analytics
into restaurant management practices has become essential for staying competitive,
responsive, and aligned with customer expectations in the digital era.

Research has shown that a half-star increase in ratings can boost peak-hour patronage by up
to 49%, and that 94% of diners consider online reviews when choosing where to eat. These
statistics underscore the importance of review content analysis as a tool for business growth
and risk management.

This study proposes scalable approaches to restaurant review analysis using NLP, sentiment
analysis, and aspect extraction techniques. The goal is to convert unstructured customer
feedback into structured, actionable insights that can inform menu design, staff training,
pricing, and strategic planning.

1.1 Statement of the Problem

Restaurant managers face significant difficulties in processing the overwhelming volume of

customer reviews across platforms. Most lack the time and resources to manually read every
comment, leading to a reactive approach where only recent or negative reviews receive
attention. Consequently, valuable insights embedded in broader review patterns often go
unnoticed.

Traditional review management tends to focus excessively on star ratings while overlooking
the nuanced information in written feedback. For instance, a three-star review stating “great
food but terrible parking” contains more actionable information than the rating alone
suggests. Without systems to detect recurring themes, restaurants may miss repeated
complaints.

Timeliness is another critical issue. If several customers raise concerns about a new dish or a
particular server, swift intervention is necessary to prevent further dissatisfaction. However,
many restaurants lack the tools for real-time issue detection.

4
Although various text analysis techniques exist, restaurant owners require practical guidance
on selecting appropriate methods for their specific needs. Understanding not just whether
reviews are positive or negative, but also what customers are specifically referring to, remains
a complex task.

This research addresses these challenges by developing strategies to extract key aspects of
dining experiences from reviews, including food quality, service interactions, ambiance,
value perception, and overall satisfaction.

Further complexities include:

● Mixed Sentiments: A single review may contain both praise and criticism, which
requires advanced techniques to parse accurately.

● Informal Language: Slang, abbreviations, and context-specific expressions can

hinder accurate interpretation.

● Demographic Variation: Different customer segments (e.g., families vs. business

clients) prioritize different elements of the dining experience.

● Temporal Factors: Changes in sentiment may reflect external events like seasonal
menu updates, renovations, or staff turnover.

● Competitive Context: Benchmarking against similar establishments adds value but

requires advanced data collection and analysis.

● Review Credibility: Some reviews may be biased, exaggerated, or fraudulent.

Differentiating authentic feedback from noise is essential for meaningful insights.

1.2 Aim and Objectives of the Study

Aim

To develop practical, scalable methods for analyzing restaurant reviews using machine
learning and NLP techniques, enabling restaurant owners to better understand customer
expectations and make targeted operational improvements.

Objectives

5
 1. To identify key aspects of the dining experience that significantly influence customer
satisfaction (e.g., food, service, ambiance, pricing).

2. To develop an automated system for extracting meaningful patterns from textual

reviews using NLP and sentiment analysis.

3. To provide actionable recommendations for restaurant owners based on review

insights.

4. To evaluate the reliability of different review platforms and develop strategies for
detecting biased or fraudulent reviews.

5. To create frameworks for analyzing sentiment trends over time, reflecting the impact
of operational changes and market shifts.

6. To identify domain-specific linguistic patterns in restaurant reviews for more accurate

interpretation.

7. To segment review analysis by customer type (e.g., families, solo diners, business
clients) to reveal distinct expectations.

8. To establish methods for competitive benchmarking across similar restaurants.

9. To determine best practices for responding to reviews, optimizing customer

engagement and reputation management.

10. To correlate review data with operational metrics (e.g., revenue, retention) to measure
business impact.

1.4 Scope of the Study

This research focuses on written reviews from major platforms (Yelp, Google Reviews,
TripAdvisor), covering a variety of restaurant types—from fast food to fine dining. It
analyzes review content related to food, service, ambiance, pricing, and overall customer
satisfaction.

The study specifically integrates machine learning (ML) models and natural language
processing (NLP) techniques to process and interpret unstructured review data. By

6
leveraging ML algorithms such as Logistic Regression, Random Forest, and Support
Vector Machines (SVM), the research aims to classify sentiments, identify recurring topics,
and extract actionable insights from textual feedback.

The scope includes:

● Reviews from the past three years, encompassing pre-pandemic, pandemic, and
recovery phases.
● English-language reviews, with future adaptability for multilingual processing.
● Both quantitative (e.g., star ratings) and qualitative (e.g., review comments) review
components.
● Implementation of ML models for classification, clustering, and pattern detection.
● Analytical techniques including sentiment analysis, topic modeling (e.g., LDA),
aspect-based sentiment analysis, keyword extraction (e.g., TF-IDF), and visualization
methods.
● Evaluation of the impact of management responses on online reputation.
● Scalable implementation strategies for restaurants with varying levels of technical
resources.

1.5 Definition of Terms

● Restaurant Review Analysis: The process of extracting insights from customer

feedback to inform restaurant operations.

● Sentiment Analysis: Identifying emotional tone (positive, negative, neutral) in

textual content.

● Text Analysis: Computational interpretation of written language.

● Topic Identification: Recognizing recurring themes within reviews.

● Aspect-Based Analysis: Focusing on specific components of a review (e.g., food,

service).

● Customer Satisfaction: The degree to which customer expectations are met.

7
● Random Forest: An ensemble machine learning algorithm that combines multiple
decision trees to improve prediction accuracy in tasks like sentiment classification.

● Keyword Extraction: Identifying high-impact words in review content.

● Logistic Regression: A supervised learning algorithm used for binary or multi-class

classification tasks, such as classifying customer reviews as positive, neutral, or
negative.

● Lexical Analysis: Processing words and symbols into meaningful units.

● Support Vector Machine (SVM): A supervised learning algorithm that works well
for text classification by finding the optimal boundary between classes (e.g., satisfied
vs dissatisfied customers).

● Clustering (e.g., K-Means): An unsupervised ML technique used to group similar

reviews or customers based on textual features, helping uncover patterns without
labeled data.

● Semantic Analysis: Understanding contextual meanings of words or phrases.

● Natural Language Processing (NLP): The field of AI focused on understanding

human language through computers.

8
 CHAPTER TWO

2.1 Literature Review

"Restaurant review analysis" refers to the systematic examination of customer feedback about
dining experiences to extract meaningful insights. Restaurant review analysis is used to
identify patterns in customer feedback, understand sentiment trends, and uncover specific
aspects of the dining experience that influence customer satisfaction. This process helps
restaurants to plan targeted improvement strategies to enhance customer satisfaction and
loyalty, which in turn reduces customer attrition and increases profitability (Rahman & Lee,
2022). Customer feedback is a critical concern for the restaurant industry, and it is
particularly significant in highly competitive markets. Negative reviews lead to decreased
patronage and financial losses while increasing the need to attract new customers (Chen,
2023). Customer retention is crucial in the restaurant business as acquiring new customers is
often five times more costly than keeping current ones. Due to the subjective nature of dining
experiences, it is quite challenging to accurately interpret customer feedback and predict
future behavior. For restaurants, it is even more complex to identify patterns in reviews due
to the unstructured nature of the data compared to other domains. This requires sophisticated
analysis techniques for effective interpretation.

The economic value of restaurant review analysis is widely recognized (Johnson & Martinez,
2021):

● Successful review analysis allows restaurants to focus more on addressing the specific
needs of their existing customers instead of constantly seeking new and potentially
unpredictable ones.
● Loyal customers are more beneficial and, if satisfied, often provide new referrals
through word-of-mouth.
● Regular customers tend to be less price-sensitive and more forgiving of occasional
service issues.
● Repeat customers become more profitable to serve due to the restaurant's knowledge
of their preferences.
● Negative reviews lead to reduced patronage, and increased marketing costs to attract
replacement customers.

9
Restaurant review analysis has become a major challenge in the food service industry, and
restaurants have increasingly tried to monitor online feedback so that they can detect
potential issues before they significantly impact business. Review analysis primarily focuses
on identifying recurring themes and sentiment patterns so that restaurants can take steps to
address concerns and enhance positive aspects (Garcia & Thompson, 2021). In today's
competitive marketplace, more and more restaurants realize that their most valuable asset is
their reputation and the data generated through customer feedback. We mainly investigate the
predictors of customer satisfaction as part of restaurant experience management (REM).
Review management is an important task to maintain and enhance restaurant reputation. Food
service establishments, such as casual dining restaurants, fine dining establishments, and
other service providers are training their staff to be more customer-oriented and they are
implementing strategies to ensure positive dining experiences (Williams & Rodriguez, 2022).
The best core marketing strategy for the future is to leverage positive reviews and prevent
negative feedback.

Previous research indicates that there were two types of targeted approaches to managing
restaurant reviews: reactive and proactive. In a reactive approach, the restaurant responds to
reviews after they are posted, addressing concerns and thanking customers for feedback. In a
proactive approach, the restaurant tries to identify potential issues before they result in
negative reviews. The establishment then tries to enhance the dining experience by providing
exceptional service and addressing concerns during the meal. If review analysis is inaccurate
then restaurants will waste their resources on unnecessary changes while missing critical
improvement opportunities, so the review analysis should be accurate.

2.1.1 Origin and Evolution of Restaurant Review Analysis Systems

Early Approaches to Restaurant Review Analysis

The concept of restaurant feedback dates back centuries, with word-of-mouth

recommendations being the primary form of review. However, formalized review analysis
gained prominence in the late 20th century with the rise of professional food critics and
published restaurant guides.

In the 1980s and 1990s, restaurants primarily relied on comment cards, guest books, and
professional critic reviews to understand customer satisfaction. Early review analysis was

10
based on manual tabulation, simple categorization, and basic statistical techniques. These
approaches provided limited analytical power, as they could not efficiently process large
volumes of feedback or capture complex patterns in customer sentiment.

During this period, the most commonly used methods included:

● Manual Content Analysis: Staff would read through comment cards and categorize
feedback.
● Sampling Techniques: Randomly selecting a subset of feedback to analyze due to
volume constraints.
● Frequency Analysis: Counting occurrences of specific comments or complaints.

Although these methods offered some insights, they lacked scalability and sophistication in
identifying subtle patterns or emerging trends in customer feedback.

The Rise of Data-Driven Review Analysis (2000s - 2010s)

The early 2000s marked a significant shift in restaurant review analysis due to the increasing
popularity of online review platforms and advancements in text analysis capabilities. The rise
of sites like Yelp (founded 2004), TripAdvisor, and later Google Reviews enabled customers
to share their dining experiences widely, creating vast repositories of restaurant feedback
data.

During this period, restaurants moved from manual analysis to more sophisticated
computational techniques, which could process larger volumes of text and extract meaningful
patterns. Some of the key techniques introduced included:

● Basic Sentiment Analysis: Identifying positive and negative language in reviews.

● Keyword Extraction: Identifying frequently mentioned terms related to food, service,
and ambiance.
● Rating Analysis: Tracking trends in numerical ratings over time.
● Word Clouds: Visual representations of commonly used words in reviews.

Industries across the hospitality sector started implementing these techniques to understand
customer feedback based on online reviews. Companies like OpenTable and Zagat leveraged
digital reviews to enhance their recommendation systems and provide more comprehensive
restaurant information.

11
The AI and Deep Learning Revolution (2010s - Present)

With the advancement of natural language processing and artificial intelligence, restaurant
review analysis systems have evolved into highly sophisticated and nuanced tools. The
integration of machine learning, deep learning architectures, and real-time analytics has
revolutionized how restaurants understand and act upon customer feedback.

Modern review analysis techniques include:

● Advanced Sentiment Analysis: Can detect nuanced emotions, sarcasm, and context-
specific sentiments in reviews.
● Aspect-Based Sentiment Analysis: Identifies specific aspects of the dining experience
(food, service, ambiance, value) and the associated sentiment for each.
● Topic Modeling: Uses techniques like Latent Dirichlet Allocation (LDA) to
automatically discover themes in reviews.
● Named Entity Recognition: Identifies specific menu items, staff positions, or
restaurant features mentioned in reviews.

The introduction of real-time review monitoring systems has enabled restaurants to take
immediate action on feedback, addressing concerns promptly and leveraging positive reviews
for marketing. Companies such as Yelp, Google, and specialized hospitality platforms now
offer AI-powered review analysis dashboards to help restaurants optimize their operations
based on customer feedback.

2.2 Machine Learning

Machine learning (ML) has transformed restaurant review analysis by providing sophisticated
algorithms that can process vast amounts of unstructured text data, detect sentiment patterns,
and improve analytical accuracy. ML-driven review analysis systems help restaurants
effectively interpret customer feedback and implement targeted improvement strategies.

2.2.1 The Role of Machine Learning in Restaurant Review Analysis

Machine learning enables restaurants to leverage large datasets of customer reviews and
uncover complex patterns in dining preferences and experiences that were previously
undetectable by manual methods. Unlike conventional review analysis techniques that rely on

12
basic keyword searches or simple rating averages, ML models automatically learn from
historical review data, continuously improve, and adapt to evolving customer expectations
and language patterns.

Key roles of machine learning in review analysis include:

● Sentiment Extraction: ML algorithms analyze review text to determine whether

customers express positive, negative, or neutral opinions about their dining
experience.
● Aspect Identification: Advanced models can identify specific aspects of the restaurant
experience mentioned in reviews, such as food quality, service speed, ambiance,
value, and cleanliness.
● Trend Detection: ML systems can track changes in customer sentiment over time,
helping restaurants identify emerging issues or improvements in specific areas.
● Competitive Benchmarking: By analyzing reviews across multiple restaurants, ML
can provide comparative insights on how an establishment performs relative to
competitors.

2.2.2 Machine Learning Techniques Used in Review Analysis

Several ML techniques are commonly used in restaurant review analysis, each offering
unique advantages in processing and interpreting customer feedback. The following are some
of the most effective approaches:

2.2.2.1 Supervised Learning Models

Supervised learning models rely on labeled datasets where sentiment and aspects have been
previously identified. These models are trained using pre-classified reviews and then applied
to new data for analysis.

a. Naïve Bayes: One of the simplest machine learning models used for text classification and
sentiment analysis. This probabilistic model is particularly effective for review classification
due to its ability to handle large vocabularies but may struggle with context and negation.

b. Support Vector Machines (SVM): This model classifies reviews by finding the optimal
decision boundary between positive and negative sentiment. It works well for text
classification tasks and is relatively robust against overfitting.

13
c. Random Forest: An ensemble learning technique that combines multiple decision trees to
improve classification accuracy of reviews. It can handle the high dimensionality of text data
and is less prone to overfitting than individual decision trees.

d. Logistic Regression: Provides a probability-based approach to sentiment classification that

offers good interpretability but may miss complex language patterns in reviews.

2.2.2.2 Unsupervised Learning Models

Unsupervised learning models are useful for discovering patterns in review data without
predefined labels, making them effective in exploratory analysis and topic discovery.

a. Latent Dirichlet Allocation (LDA): LDA is a topic modeling technique that groups reviews
based on similar themes, enabling restaurants to identify common topics mentioned across
multiple reviews without manual categorization.

b. K-Means Clustering: K-Means clustering groups similar reviews together, helping

restaurants identify distinct categories of customer experiences and feedback patterns that
might require different response strategies.

2.2.2.3 Deep Learning Techniques

Deep learning models have become increasingly popular for review analysis due to their
ability to understand context, capture semantic meaning, and process complex language
patterns.

a. Word Embeddings (Word2Vec, GloVe): These techniques represent words as vectors in a

continuous space, capturing semantic relationships between terms used in restaurant reviews
and improving the performance of other ML models.

b. Convolutional Neural Networks (CNN): CNNs can identify local patterns in text, making
them effective for detecting specific phrases or expressions that indicate customer sentiment
toward particular aspects of the dining experience.

c. Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM): These models
are particularly effective for analyzing the sequential nature of language in reviews, capturing
context and dependencies between words to better understand customer sentiment.

14
d. BERT (Bidirectional Encoder Representations from Transformers): This advanced NLP
model has revolutionized review analysis by understanding context in both directions,
allowing for more accurate interpretation of complex language features like sarcasm,
comparative statements, and nuanced opinions.

2.2.3 Advantages of Machine Learning Over Traditional Review Analysis Methods

Feature Traditional Methods Machine Learning Models

Accuracy Lower due to subjective Higher due to statistical pattern

interpretation recognition

Scalability Struggles with large review Handles thousands of reviews

volumes efficiently

Consistency Varies based on human Consistent application of learned

analyzer patterns

Processing Speed Slow, manual processing Fast, automated analysis

Detail Level Often misses subtle patterns Captures nuanced sentiment and
specific aspects

Language Typically limited to one Can be trained on multiple languages

Capability language

2.3 Challenges of Traditional Approaches to Restaurant Review Analysis

Understanding customer feedback is crucial for restaurants seeking to improve service, menu
offerings, and overall customer satisfaction. Traditionally, many restaurants have relied on
manual or basic methods to interpret reviews — such as keyword spotting, star rating
averages, or periodic sentiment audits. While these approaches offer some insight, they fall
short when it comes to extracting meaningful, real-time, and actionable insights from large
volumes of unstructured data. This section outlines the core limitations of conventional
methods in restaurant review analysis and underscores the benefits of transitioning to
machine learning-based techniques.

15
1. Low Precision in Sentiment Interpretation

Traditional techniques often lack the depth required to accurately interpret the nuances of
customer sentiments in reviews. These methods typically depend on straightforward metrics
like word frequency or predefined sentiment dictionaries. However, such tools struggle with:

● Misinterpreting Context: Words can carry different meanings depending on how

they are used. For instance, a phrase like “hot service” could be either a compliment
or a complaint depending on context.

● Oversimplified Sentiment Scoring: Assigning a general tone to entire reviews

overlooks mixed or subtle opinions that are essential for targeted improvements.

● Limited Attention to Review Structure: Traditional systems rarely consider

sentence relationships or paragraph flow, which are often key to understanding a
reviewer’s full experience.

Due to these shortcomings, restaurants using basic techniques risk drawing incorrect
conclusions about customer satisfaction levels and priorities.

2. Inability to Keep Up with Changing Language and Trends

Customer language evolves quickly, especially online where slang, emojis, abbreviations, and
cultural references are constantly shifting. Static models based on fixed rules or outdated
lexicons face the following issues:

● Lack of Flexibility: Once established, rule-based systems rarely evolve, making them
ineffective against new expressions or review styles.

● No Learning Mechanism: Without the ability to adapt from incoming data, these
systems become stale and less relevant over time.

● Blindness to Emerging Topics: Traditional models are often blind to trending

themes in feedback, such as sudden dissatisfaction with a new menu item or staff
change.

16
As a result, restaurants may miss early warning signs of operational issues or customer
dissatisfaction that could be addressed proactively.

3. Reliance on Structured and Manually Filtered Data

Most traditional review analysis tools focus solely on structured data like star ratings, survey
responses, or tagged complaint types. This approach is limited in several ways:

● Neglect of Textual Data: Unstructured feedback — such as detailed customer

reviews or social media comments — holds valuable insights but is often ignored or
underused.

● Manual Tagging Burden: Conventional methods require human intervention to

classify themes or topics, which is time-consuming and prone to error.

● Siloed Data Sources: Feedback may be spread across different platforms (e.g.,
Google, Yelp, Instagram), and traditional systems often lack the capacity to unify
these sources.

By failing to harness the full spectrum of available data, restaurants risk overlooking the root
causes of customer dissatisfaction.

4. High Rate of Misclassification

Basic review analysis methods often mislabel feedback, leading to skewed results and poor
decision-making. Two common types of misclassification include:

● Overestimating Negativity: A review that mentions “waited a bit long, but the food
was amazing” may be flagged as negative simply due to the word “waited.”

● Missing Constructive Feedback: Comments that contain suggestions but no clear

praise or complaint may be ignored altogether.

These inaccuracies can cause restaurants to take unnecessary actions or overlook

opportunities for improvement.

17
5. Lack of Timely and Personalized Insights

In a competitive food service environment, responding to customer reviews quickly and

effectively can make a significant difference. Traditional systems often fall behind due to:

● Delayed Processing: Many rely on periodic review rather than continuous analysis,
which means critical issues may go unnoticed for days or weeks.

● Generalized Responses: Without real-time analysis, restaurants may send generic

replies or apply broad changes that don’t address specific concerns.

● Missed Engagement Opportunities: Timely recognition of feedback allows for

meaningful interaction — such as apologizing, thanking, or offering incentives —
which static systems rarely support.

2.4 Research Gap

Although numerous studies have been conducted on the analysis of restaurant reviews using
various machine learning and natural language processing (NLP) techniques, several
important gaps still exist. While tools like sentiment analysis, topic modeling, and aspect-
based sentiment analysis (ABSA) have been widely applied, many of these works either
oversimplify the complexity of customer feedback or fail to adapt effectively to the dynamic
nature of online reviews.

A major concern is that many models are trained on generic or benchmark datasets, such as
Yelp or IMDB, without adequately addressing the domain-specific language used in
restaurant reviews. These reviews often contain informal expressions, slang, emojis, spelling
errors, and abbreviations that significantly affect text understanding. As such, models not
tailored to these specific patterns may produce misleading sentiment interpretations.

Furthermore, a lot of attention has been placed on classifying reviews as positive, negative, or
neutral. However, these binary or ternary sentiment categories often overlook the richer
context embedded in customer feedback. For instance, a review might praise the food quality
but criticize the service, leading to mixed sentiments within the same statement. Many
current approaches struggle to interpret such nuances effectively.

18
Despite the emergence of more powerful language models, including BERT, RoBERTa, and
GPT-based architectures, there is limited application of these models in restaurant review
analysis. The few studies that do incorporate them often do not fine-tune the models
specifically for the restaurant or hospitality domain, thereby missing opportunities for
improved performance. Additionally, these models typically require significant
computational resources, which limits their adoption in small businesses or low-resource
environments.

Finally, existing research tends to focus heavily on review texts from a single platform.
However, in practice, customers leave feedback across multiple platforms such as Google
Reviews, TripAdvisor, Facebook, and Instagram. The failure to aggregate and analyze cross-
platform feedback limits the depth and reliability of insights generated. Moreover, there is
minimal exploration of real-time sentiment analysis tools that could help restaurant managers
make immediate, data-driven decisions to improve service delivery.

2.4.1 Lack of Domain-Specific Sentiment Modeling

One recurring issue across the reviewed works is the application of generalized models to
domain-specific tasks. Restaurant reviews contain unique linguistic patterns, such as culinary
terms, local jargon, or customer-service-related expressions. When models are not
customized to recognize these terms, the accuracy of sentiment classification declines. There
is a clear need for models that are pre-trained or fine-tuned on restaurant-specific datasets to
achieve better contextual understanding.

2.4.2 Limited Focus on Multi-Aspect Sentiment Detection

Another gap is the insufficient attention paid to detecting and analyzing multiple aspects
within a single review. Many systems process each review as a single unit of sentiment
without separating feedback related to food, ambience, service, pricing, and cleanliness. This
leads to shallow insights and a poor understanding of what specifically influences customer
satisfaction or dissatisfaction. Although some ABSA models exist, they still struggle with
ambiguity and implicit references to aspects.

2.4.3 Inadequate Use of Modern Deep Learning and Transformer Models

19
Although modern transformer-based models like BERT and GPT have revolutionized NLP,
they are not yet widely used in this specific domain. Several studies continue to rely on
traditional classifiers such as Support Vector Machines (SVM), Naive Bayes, or even basic
neural networks. This limits the potential of achieving state-of-the-art accuracy and
understanding. Furthermore, when deep learning is used, it is rarely integrated with attention
mechanisms or contextual embeddings that could provide deeper sentiment interpretation.

2.4.4 Need for Real-Time and Cross-Platform Review Analysis

There is also a significant gap in real-time analytics and integration of sentiment insights into
operational systems. Most studies analyze static data and fail to show how insights can be
applied in a timely manner to resolve customer complaints or optimize services. Moreover,
customers use diverse platforms to leave feedback, and studies that analyze only one source
are not representative of overall customer sentiment. Merging data from multiple platforms
and handling the challenges of heterogeneity remains an open research problem.

2.5 Related Works

No Author Journal and Year Methodology Challenges

1 Michael Journal of AI Sentiment analysis using Difficulty handling

Johnson, Research, 2019 NLP techniques to sarcasm and mixed
Emily Reed classify reviews as sentiment in reviews.
positive or negative.

2 Laura Computational Hybrid model combining High variance in text

Taylor, Linguistics, 2020 CNN and LSTM for length and diverse
James sentiment prediction in customer language
Morgan restaurant reviews. styles.

3 Amanda Data Mining and Aspect-based sentiment Difficulty in handling

Williams, Knowledge analysis to analyze domain-specific
Richard Discovery, 2021 reviews based on terminology and non-

20
 Lewis restaurant features (e.g., standard expressions.
food quality, service).

4 Benjamin Machine Learning Comparison of machine Inconsistent labeling

Brown, Review, 2020 learning models: Naive and noisy data leading
Chloe Bayes, SVM, and to lower accuracy.
Adams Random Forest.

5 Rachel Journal of Natural Deep learning model High computational

White, Language using BERT for cost of training BERT
Daniel Clark Processing, 2021 sentiment classification on large datasets.
in restaurant reviews.

6 Nathan Artificial Hybrid sentiment Ambiguities in

Harris, Intelligence analysis approach sentiment
Olivia Green Journal, 2020 combining lexicon-based classification and the
methods and machine need for labeled
learning algorithms. datasets.

7 Ethan IEEE Transactions Sentiment and aspect- Difficulty in accurate

Walker, on AI, 2021 level analysis to identify aspect extraction due
Megan features influencing to ambiguous
Foster customer satisfaction. wording in reviews.

8 Christopher International Sentiment classification Balancing model

Robinson, Journal of models including Naive accuracy with
Sarah Machine Learning, Bayes and Logistic interpretability of
Bennett 2020 Regression. results.

9 Andrew Journal of Data Feature-based sentiment Challenges with

Clark, Science, 2019 classification focusing on inconsistent feature
Victoria specific restaurant identification across
Hughes attributes (e.g., reviews.
atmosphere, food
quality).

21
10 Thomas Journal of Analysis of restaurant Difficulty
Lewis, Computational reviews using machine distinguishing subtle
Grace Evans Linguistics, 2020 learning models like nuances and mixed
Random Forest and sentiments in
SVM. feedback.

11 Jennifer Journal of AI and LSTM model for Large dataset size,

Smith, Data Science, sentiment analysis of causing long
Adam Hall 2020 reviews, combined with processing times and
word embeddings for memory constraints.
better context
understanding.

12 Patrick Journal of Data Aspect-based sentiment The challenge of

Adams, Mining, 2020 analysis focusing on extracting and
Lauren categorizing review categorizing
James features into positive or ambiguous and vague
negative sentiments. aspects.

13 Sophia International Use of deep learning and Issues in

Scott, Journal of Data sentiment lexicons for distinguishing
Matthew Analytics, 2021 review analysis. context-dependent
Wright sentiment and review
quality.

14 Dylan Journal of Combination of Sparse data and lack

Miller, Computational supervised and of labeled datasets
Abigail Science, 2020 unsupervised learning made training
Turner techniques for review challenging.
sentiment extraction.

15 Lucas Davis, Journal of Evaluation of multiple Difficulty in

Samantha Business classifiers for sentiment capturing subtle
Mitchell Analytics, 2021 analysis, including sentiments and short
decision trees and SVM. review lengths.

22
16 Henry Journal of Digital Using feature extraction Problems with noisy
Martin, Marketing, 2020 and ensemble methods to data and handling
Charlotte analyze restaurant review multiple languages in
Nelson data. reviews.

17 Gabriel Journal of AI Text classification High computational

Green, Technology, 2021 models using support cost and need for
Natalie vector machines for fine-tuning the model
Simmons review sentiment for better accuracy.
analysis.

18 Jason Evans, Journal of Hybrid approach Handling varying

Lily Grant Restaurant combining LSTM with review quality and
Management, attention mechanisms to textual ambiguity in
2021 improve review feedback.
interpretation.

19 Jonathan Data Science & Comparison of various Lack of domain-

Carter, Zoe Analytics Journal, machine learning specific datasets and
Phillips 2020 techniques for restaurant generalizability
review analysis. issues.

20 Olivia International Multi-class sentiment Challenge in

Parker, Journal of NLP classification using accurately predicting
George Applications, 2021 XGBoost and TF-IDF neutral or mixed
Price features. reviews.

21 Isaac Bell, AI & Society Ensemble-based Inconsistent review

Hannah Journal, 2022 sentiment classifier structures and
Watson trained on diverse informal language
restaurant review styles.
sources.

22 Victoria NLP Advances, Topic modeling using Topics overlap and

Reed, Aaron 2020 LDA to identify key loss of sentiment
Campbell themes in restaurant context in clustering.

23
 feedback.

23 Samuel Applied AI Lexicon-enhanced deep Scalability issues and

Brooks, Mia Review, 2021 learning model for poor performance on
Anderson sentiment analysis. slang and
abbreviations.

24 Elijah AI-Driven Aspect term extraction Ambiguity in user

Peterson, Services Journal, combined with opinion expressions made
Ruby Hayes 2020 mining. extraction less
accurate.

25 Jack Text Analytics in Rule-based and ML- Rules were too rigid,
Bennett, Practice, 2021 based comparison for while ML required
Emily Stone restaurant review large training data.
analysis.

26 Grace International Graph-based approach to Graph complexity and

Murphy, Journal of AI and link sentiments with interpretation
Leo Russell Society, 2022 restaurant service challenges.
features.

27 Ethan Journal of AI Deep BiLSTM model Review imbalance

Brooks, Applications, 2020 trained on Yelp and (more positives than
Isabelle Zomato reviews. negatives) affected
Morris model learning.

28 Daniel King, NLP Systems & Fine-tuned RoBERTa RoBERTa required

Ava Wallace Design, 2021 model on domain- significant computing
specific restaurant review resources.
data.

29 Noah Long, Modern AI Zero-shot classification Low performance on

Ella Hart Journal, 2021 using GPT for new unseen domain-
restaurant review topics. specific features.

30 Liam Foster, Review Transformer-based Difficulty

24
 Amelia Lane Intelligence sentiment analyzer with understanding implied
Journal, 2022 attention to aspect-terms. sentiments and
sarcasm.

2.6 Synthesis of the Related Works

The reviewed studies provide a comprehensive view of how NLP and machine learning
techniques have been applied to restaurant review analysis. Sentiment analysis, aspect-based
sentiment analysis (ABSA), and topic modeling have been frequently used to extract insights
into customer opinions on restaurant services, food quality, and overall experiences. Machine
learning models such as Naive Bayes, SVM, and deep learning approaches like CNNs and
LSTMs have demonstrated strong performance in classifying and predicting sentiments.

Key challenges across the studies include handling noisy and unstructured data, such as
slang, misspellings, and sarcasm, as well as the high computational costs associated with
training deep learning models. The need for effective feature engineering and the difficulty of
capturing the diversity of customer feedback are also common challenges. Hybrid models
that combine various machine learning and NLP techniques have shown promise in
addressing some of these issues.

25