INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7181
www.rsisinternational.org
Sentiment Analysis towards Car Reviews With Data Visualization
Mohamad Hafiz Khairuddin*, Nurazian Binti Mior Dahalan, Albin Lemuel Kushan, Nur Farhana Binti
Mohd Nasir
Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA (UiTM) Cawangan
Melaka Kampus Jasin, 77300 Merlimau, Melaka
DOI: https://dx.doi.org/10.47772/IJRISS.2025.910000585
Received: 27 October 2025; Accepted: 02 November 2025; Published: 19 November 2025
ABSTRACT
Nowadays, there are too many car reviews on the internet, worldwide. Big manufacturing companies use user
feedback to improve product quality by understanding the user perspective. Customers will read reviews on
websites or on social media platforms before deciding which cars to buy, and may consider testing at a nearby
showroom. So, reviews are very important to both the manufacturer and the customer. Nevertheless, it is hard
to extract useful information from hundreds or thousands of reviews on websites or social media platforms.
Sentiment analysis is applied across various areas, such as business and products, to analyse and learn from
people’s opinions. Fine-grained sentiment analysis is best for analysing the polarity of a sentence and
determining its sentiment —positive, negative, or neutral. After preprocessing the reviews, extract features and
use Naïve Bayes to classify sentiment. The results will be displayed in the dashboard visualisation so the user
can read all the reviews properly. Functional testing is conducted to ensure the system runs smoothly, as it
should. There is a need to improve this system, as some of these car models are not very common in Malaysia.
Later, we can get data on Malaysia's standard car models and apply the system to them. The model's
classification method accuracy could be improved by training and testing the system on a large number of
reviews.
Keywords: Car Review, Sentiment Analysis, Naïve Bayes
INTRODUCTION
With the ongoing economic development, more and more families are considering buying a car. For ordinary
families, buying a car is an important and relatively expensive thing. So, it is important to choose a car which
has a suitable price and quality. Examples of top automobile companies are Audi, BMW, Honda, and
Mitsubishi. Users have many choices when it comes to their favourite automobile companies, but most people
prefer those that offer high-quality products and good service. With that, a high-quality product can be an
important service. Quality is defined as meeting or exceeding the client’s expectations.
TikTok, Facebook, Google, and other online platforms like Shopee are now among the most popular platforms
for users to share their buying experiences through online reviews and ratings, or to share important
information for users. (Adwan et al., 2020). On many of these platforms, buyers can share their feelings,
opinions, or even suggestions about the product or the manufacturer. For example, users post their comments
on online websites that are easy to access and usually include a star rating ranging from 1 to 5 (Alamanda et
al., 2019). So, users no longer refer to others when buying a car, as they can read reviews on websites. In
particular, while users read reviews, ratings do not always accurately reflect their review sentiment.
Furthermore, the available data in raw format is not an easy task to analyse in a limited time. The reviews
could be a challenge for a user to extract important information, as they include too many types of information
in a single sentence. Therefore, a study suggests that sentiment analysis is needed, as it can help classify
reviews into positive and negative categories to improve service performance and product quality (Panchal &
Deshmukh, 2020).
Sentiment analysis (also referred to as subjectivity analysis, opinion mining, or emotion artificial intelligence)
is a natural language processing (NLP) technique that identifies patterns and features from a large text corpus
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7182
www.rsisinternational.org
(Lamba, Manika, & Margam; Madhusudhan, 2022). Negative and slightly negative ratings frequently result in
sales loss. In addition, machine learning, lexicon-based, and aspect-based sentiment analysis are practical
approaches that can produce categorical sentiment (positive or negative).
A Naïve Bayes classifier can classify Twitter tweets into positive or negative (Al-Natour & Turetken, 2020). It
also operates on the principles of probability and assumes independence among features, which is why it is
called “naïve”. For sentiment analysis, it can effectively handle textual data considering the presence of certain
words or phrases (Deshmukh et al., 2023). The dashboard visualisation highlights sentiment analysis results to
help the user better understand them.
Problem Statement
Nowadays, customers do not have to enter the shop to extract the information about the automobile they wish
to purchase. They can get all kinds of information immediately by clicking a mouse and browsing social media
platforms. However, with easy access to information, this creates a disequilibrium between demand and supply
(Shamsher Singh & Ameet Sao, 2021). So, people just believe reviews on any social media platform, but not
on the official website. Every car buyer in the country starts their search on the World Wide Web. Social media
platforms have become common channels for businesses and organisations to market their products. Extracting
meaningful information from consumer reviews, such as the most frequent words and their relationships,
provides the company with insights to address and resolve issues quickly (Kim, E., & Chun, S., 2019).
So, consumers face challenges in choosing the right cars from a vast network due to the sheer volume of data,
diverse types, and the low density of valuable information. Collecting consumer feedback through online
surveys is both expensive and time-consuming for automobile companies (Panchal & Deshmukh, 2020). This
is because reviews may not depict the quality of the company's products. Consequently, automobile companies
find it challenging to deliver service or product quality that surpasses consumer expectations, leading to a
direct loss of both potential customers and revenue.
Thus, Awais et al. (2020) suggest that companies should collect user experience data and perform sentiment
analysis to assess the polarity of the text —whether it is a positive or negative review. Extracting insights from
such feedback can contribute to knowledge. When the company analyses the polarity of reviews, it can gauge
user opinions on the quality and effectiveness of its products.
Related Works
There are three types of related work similar to this project: Machine Learning Model for Sentimental Analysis
of Amazon Reviews, Sentiment Analysis for social media using SVM Classifier, and Sentiment Analysis of
YouTube Movie Trailer Comments using Naïve Bayes.
Machine Learning Model for Sentimental Analysis of Amazon Reviews
This research aims to deepen the understanding of online product reviews by examining a large Amazon
dataset comprising numerous star ratings and comments (Umamageswari et al., 2024). The motivation for this
project is that consumers currently use product reviews as a decision-making tool when buying. It represents
the quality and dependability of the products. So, this research aims to ensure that ratings and reviews are
correlated, not vice versa. The research used three models: Random Forest, Gradient Boosting, and a Hybrid
(Random Forest & Gradient Boosting). The accuracy results for Random Forest are 73%, while for Random
Forest are 76%. The hybrid model shows 92%. So, it is concluded that a combination of individual classifiers
can outperform a single effective classifier.
Sentiment Analysis for social media using SVM Classifier of Machine Learning
This research shows the significance of doing sentiment analysis for businesses and organisations. Support
Vector Machines (SVMs) are machine learning techniques used for sentiment analysis (Huang, 2023). The
research found that sentiment analysis using SVM has been proven to be a practical approach for analysing
social media data in business and organisations. The research is focusing on sentiment analysis of US-Airlines-
related tweets. The precision, recall, and F1-score indicate that SVM is a promising approach for sentiment
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7183
www.rsisinternational.org
analysis and text polarity classification. With these valuable findings, businesses and organisations can better
manage social media sentiment and make more informed decisions. The accuracy for this project is 91.8%.
The precision is 91.3%, and the recall is 82.3%. The F1-score is 86.9, and these are the results of analysing
US-Airlines related tweets by the SVM algorithm.
Sentiment Analysis of YouTube Movie Trailer Comments using Naïve Bayes
This research analyses viewers’ comments and opinions on YouTube about Money Heist, a Netflix TV series
(Novendri et al., 2020). There are four seasons of Money Heist. However, many still comment neutrally or
give positive feedback about the series. So, sentiment analysis is conducted to classify opinions using Naïve
Bayes. It is chosen because a previous study showed satisfactory results. The Naïve Bayes algorithm has been
used in text mining because it is simple yet can achieve high accuracy. The result is considered successful
because the accuracy rate is 81%. The precision is 74.83%, and the recall is 75.22%.
METHODOLOGY
This phase involved tasks such as Project Framework, Use Case Diagram, and Flow Chart.
Project Framework
Figure 1 illustrates the phases of the project framework for system development. In here, it shows the sequence
of each phase along with the activities, as well as the expected outcomes after the development is completed
Fig 1. Project Framework
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7184
www.rsisinternational.org
Use Case Diagram
A use case diagram is designed and created to meet the requirements in the design phase of the Modified
Waterfall Methodology. A use case can show the interaction between the user and the system. It can also see
how many tasks are needed to use the system. Figure 2 shows the use case diagram.
Fig 2. Use Case Diagram
Flowchart
A flowchart is a diagram that shows how data moves through an organisation. It gives a clear picture of the
actions taken and the order in which they are carried out within a system. The system flowchart is shown in
Figure 3.
Fig 3. Flowchart
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7185
www.rsisinternational.org
RESULTS
In an accuracy test, all the processed data has been split into 20:80, with 20% used for testing and the
remaining 80% for training. This ratio is chosen because it yields the highest accuracy score among the other
ratios and has already been tested. The technique used is called the Confusion Matrix to measure and enhance
the performance of the developed model. Figure 4 shows the result of the Confusion Matrix by percentage.
Fig 4. Confusion Matrix in Percentage
In Figure 5.1, the rows of the matrix correspond to accurate labels for positive, negative, and neutral sentiment.
The columns for the predicted labels of positive, negative, and neutral. It is presented in Table I in 4 parts: True
Positive, True Negative, False Positive, and False Negative.
Table I Classifier Parameters
True / Predicted
Negative
Neutral
Positive
Negative
87.74
1.04
11.21
Neutral
10.13
83.98
5.90
Positive
12.75
2.00
85.24
True Negative (TN)
The true class is Negative, and the classifier correctly predicted Negative with 87.74.
False Positive (FP)
The true class is Negative, but the classifier predicted either Neutral or Positive. It is 1.04 as (Neutral) and
11.21 as (Positive).
False Negative (FN)
The true class is Neutral or Positive, but the classifier predicted as Negative. It is 10.13 (true Neutral) and
12.75 (true Positive).
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7186
www.rsisinternational.org
True Positive (TP)
The true class is either Neutral or Positive, and the classifier correctly predicted either Neutral or Positive. It is
83.98 (actual Neutral) + 5.90 (true Positive) + 85.24 (true Positive).
Figure 5 shows the evaluation results for the Naïve Bayes model.
Fig 5. Naïve Bayes Model Evaluation Results
CONCLUSION
In conclusion, several functions have been recognised during project development. This system utilised the
Naïve Bayes model to categorise the reviews into polarities. The analysis results are then displayed in a
dashboard using Anvil Editor, an open-source app framework. It is also convenient for designing a simple
interface. Several limitations have also been identified during the project's development and testing phases.
The project relies on inadequately labelled data, where the abundance of positive reviews relative to negative
and neutral ones could affect the classifier’s performance. The system focuses only on English reviews
because the dataset used in this system development comes from overseas users. So, it might not perform very
well with other languages. The project takes around 25 seconds to load the visualisation graph when the user
clicks the link because the image is quite large. Therefore, a few recommendations for future improvements
are required. Utilising a properly labelled dataset that maintains a balance between positive and negative
instances is recommended for this project. When the dataset is balanced, it can improve the classifier's
performance. The system could be more useful if it recognised other languages, since the world is a diverse
place with many people, races, and languages. The system can be upgraded by loading the graph image when
the user clicks, or by loading all when the user clicks the visualisation link. Despite the limitation, it meets all
the project objectives: to design a machine learning model and a dashboard visualisation for sentiment analysis
of car reviews, to develop a web-based platform for sentiment analysis with data visualisation, and to test the
effectiveness of the web-based system.
REFERENCES
1. Adwan, O. Y., Al-Tawil, M., Huneiti, A. M., Shahin, R. A., Abu Zayed, A. A., & Al- Dibsi, R. H.
(2020). Twitter sentiment analysis approaches: A survey. International Journal of Emerging
Technologies in Learning, 15(15), 79–93.
2. Alamanda, D. T., Ramdhani, A., Kania, I., Susilawati, W., & Hadi, E. S. (2019). Sentiment analysis
using text mining of Indonesia Tourism reviews via social media. International Journal of Humanities,
Arts and Social Sciences, 5(2), 72– 82.
3. Al-Natour, S., & Turetken, O. (2020). A comparative assessment of sentiment analysis and star ratings
for consumer reviews. International Journal of Information Management.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025
Page 7187
www.rsisinternational.org
4. Awais, M., Batool, S., Mirza, A. M., Sajid, A., Khokhar, A. S., & Zafar, A. (2020). Patient’s feedback
platform for quality of services via “Free Text Analysis” in healthcare industry. EMITTER
International Journal of Engineering Technology, 8(2), 316–325.
5. Deshmukh, A., Sonar, S. D. B., Ingole, R. V., Agrawal, R., Dhule, C., & Morris, N. C. (2023). Satellite
image segmentation for forest fire risk detection using Gaussian mixture models. In Proceedings of the
2023 2nd International Conference on Applied Artificial Intelligence and Computing (ICAAIC) (pp.
806–811). IEEE.
6. Huang, Q. (2023). Sentiment analysis for social media using SVM classifier of machine learning.
Applied and Computational Engineering, 4(1), 86–90. https://doi.org/10.54254/2755-2721/4/20230354
7. Kim, E., & Chun, S. (2019). Analyzing online car reviews using text mining. Sustainability, 11(6),
1611. https://doi.org/10.3390/su11061611
8. Lamba, Manika & Margam, Madhusudhan. (2022). Sentiment Analysis. 10.1007/978- 3- 030-85085-
2_7.
9. Novendri, R., Callista, A. S.., Pratama, D. N., & Puspita, C. E. (2020). Sentiment Analysis of YouTube
Movie Trailer Comments Using Naïve Bayes. Bulletin of Computer Science and Electrical
Engineering, 1(1), 26–32. https://doi.org/10.25008/bcsee.v1i1.5
10. Panchal, D. S., Kawathekar, S. S., & Deshmukh, S. N. (2020). Sentiment analysis of healthcare quality.
International Journal of Innovative Technology and Exploring Engineering, 9(3), 3369–3376
11. Singh, S., & Sao, A. (2021). Impact of social media marketing in consumer buying behavior in
automobile industry: an empirical study in Delhi. Turkish Online Journal of Qualitative Inquiry, 12(7),
6278–6292. https://tojqi.net/index.php/journal/article/view/4832
12. Umamageswari, A., Pratishwaran, R. J., Reddy, M. P., & Raj, R. Y. S. (2024). Machine learning model
for sentiment analysis of Amazon reviews. In Proceedings of the 2024 International Conference on
Electronic Systems and Intelligent Computing (ICESIC) (pp. 139–144). IEEE.
