Sentiment Analysis in Customer Reviews for

2024 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) | 979-8-3503-8944-9/24/$31.00 ©2024 IEEE | DOI: 10.1109/ACCAI61061.2024.10602027

Product Recommendation in E-commerce Using

Machine Learning
JeidyPanduro-Ramirez

Universidad Autónoma del Perú, Lima, Perú

E-mail : [email protected]

Abstract- When it comes to delivering product consumers convey in their reviews, sentiment analysis,
recommendations on e-commerce platforms, the purpose of which is a subset of natural language processing (NLP),
this research project is to evaluate the use of sentiment has developed into a valuable tool that has gradually risen
analysis in customer reviews to give recommendations. in importance over time[2]. By conducting a study of the
During the course of this inquiry, the procedures that were
utilized were developed from various machine-learning
polarity of customer evaluations, e-commerce platforms
approaches. Customer reviews have emerged as a significant can gain substantial insights into the level of satisfaction
source of information for both customers and businesses experienced by consumers, identify areas in which further
alike as a result of the meteoric rise in the popularity of development is required, and modify product proposals to
online purchasing. This is because customers provide better meet the requirements and preferences of their
feedback on products and services that they have purchased customers[3]. It is possible to acquire these insights by
electronically. Through the utilization of sentiment analysis, analyzing the feelings that are represented in evaluations
e-commerce platforms can glean insights from the textual written by customers. How this objective is attained is the
comments they receive from customers. After these insights determination of whether the reviews are positive,
have been collected, they can be used to enhance the
algorithms that are used to recommend products to specific
negative, or neutral based on the information provided.
individuals. To analyze customer reviews, categorizing
sentiments as either positive, negative, or neutral, and In the context of online retail marketplaces, the purpose of
determining the essential characteristics and components this research project is to examine the application of
that have an impact on customer satisfaction, we employ a sentiment analysis in customer evaluations to make
strategy that involves the utilization of machine learning product suggestions. The application of machine learning
algorithms. techniques will be how this objective will be attained[4].
Extraction of meaningful insights from textual consumer
Keywords:-Sentiment analysis, Customer reviews, Product comments, identification of feelings, and incorporation of
recommendation, E-commerce, Machine learning, Online
shopping, User feedback.
these feelings into product recommendation algorithms
are all part of our approach. With this, we will be able to
provide our customers with recommendations that are
I. INTRODUCTION
more tailored to their individual needs and more pertinent
to their specific requirements. This objective will be
Reviews that are written by customers are of the utmost
accomplished by the application of machine learning
value in the rapidly growing business of e-commerce.
strategies that are of an exceptionally high level of
This is because reviews play a significant part in the
effectiveness.
decision-making process regarding purchases and have a
significant amount of influence over the behavior of
The significance of this research lies in the fact that it can
customers. As a result of the growth of online shopping
improve the efficiency of product suggestion algorithms,
platforms, consumers are increasingly relying on the
which are utilised in websites that promote online
opinions and judgments of other shoppers to guide their
shopping. This research is vital because it has the
selections regarding the things they intend to purchase[1].
potential to raise the performance of these algorithms.
This is a trend that is expected to continue. As a
When it comes to making suggestions, traditional
consequence of this, companies who engage in e-
recommendation systems typically only take into account
commerce are confronted with the task of exploiting the
data that is pertinent to the interaction that takes place
feedback efficiently offered by customersto improve the
between the user and the item that is being
entire shopping experience and increase sales.
recommended[5]. The user's browsing history and
purchase history are two examples of the types of
When it comes to gleaning insights from textual data and
information that could be included in this data. These
providing an understanding of the emotions that

Authorized licensed use limited to: Zhejiang University. Downloaded on January 08,2025 at 07:44:51 UTC from IEEE Xplore. Restrictions apply.
algorithms have the potential to be useful to a certain researchers have suggested should be done to improve the
extent; nevertheless, they may fail to take into accuracy and relevance of recommendations. The
consideration crucial information that is provided in the researchers have taken into consideration this idea as a
form of textual customer reviews. This is even though potential solution. Because recommendation systems can
these algorithms have the potential to be extremely better capture user preferences and provide ideas that are
valuable[6]. This information might include specific more contextually relevant when they make use of
product qualities, quality, and user experiences. There is a sentiment information that is acquired from customer
possibility that this information will include certain evaluations, they can create more effective
things. recommendations. The reason for this is that they can
determine the preferences of the user with even greater
II.RELATED WORKS precision.

Both academics and professionals working in the business When it comes to the field of sentiment analysis, there has
have seen, over the course of the past few years, that the been a significant degree of application of methods that
incorporation of sentiment analysis into product are related to machine learning tactics. Aspect-based
recommendation systems in online commerce has sentiment analysis, opinion mining strategies, and
gathered a substantial amount of interest[7]. This is sentiment categorization are some of the techniques that
something that has come to the attention of both groups. are included in this category of techniques. Researchers
The purpose of this part is to give a literature review on have investigated a variety of learning strategies,
the topic of sentiment analysis in customer reviews to including supervised learning, unsupervised learning, and
provide product recommendations in e-commerce through semi-supervised learning, to evaluate the sentiment that is
the application of techniques that are related to machine included in customer evaluations and to derive valuable
learning. This section's mission is to provide this review. insights. These strategies have been researched to assess
the sentiment that is contained in customer evaluations.
The application of methods for sentiment analysis within Deep learning architectures, such as convolutional neural
the context of online commerce has been the subject of networks (CNNs) and long short-term memory networks
several studies that have been conducted. These studies (LSTM), have demonstrated that they can capture
have been conducted to gain an understanding of the intricate correlations in textual data and attain state-of-
preferences and viewpoints of consumers by the-art performance in tasks involving sentiment analysis.
concentrating on the extraction of sentiment polarity from This is according to the findings of a recent study. We
customer evaluations. Several different machine learning have made a huge step forward in the field of deep
strategies, such as support vector machines (SVM), Naive learning with this development.
Bayes, and recurrent neural networks (RNNs), have been
investigated by researchers to develop applications that III.RESEARCH METHODOLOGY
are capable of categorizingsentiments[8]. The
categorization of sentiments has been investigated using The goal of the methodology that has been described is to
these various methodologies by researchers. According to develop a comprehensive framework for the analysis of
the findings of these studies, sentiment analysis is an sentiment in customer reviews for product suggestions in
effective method for gaining valuable insights from the online retailing by utilizing techniques from machine
feedback provided by customers, and it also can bring learning. This will be accomplished by utilizing the
about improvements in the decision-making process approaches that have been presented. To detect products
inside e-commerce platforms[9]. It has been demonstrated and make product recommendations to customers, this
through the results of these investigations that sentiment framework will be utilized. The plan that we have is to
analysis is an innovative and useful method. extract sentiment information from textual user
comments, classify the sentiments, and then include them
Customers are commonly provided with tailored product into algorithms that are used to recommend products. The
options through the usage of traditional product ability to provide clients with recommendations that are
suggestion systems, which are frequently utilized in the more tailored to their specific needs and requirements will
realm of e-commerce. Collaborative filtering and content- be made possible as a result of this. This objective will be
based filtering methodologies are frequently utilized by accomplished by the application of machine learning
these systems when it comes to the provision of strategies that are of an exceptionally high level of
recommendations for users. The previous actions and effectiveness.
preferences of the consumers serve as the foundation for
these aforementioned tactics. Even though these methods Our strategy starts with the gathering of client input from
are effective, it is quite probable that they cannot take into several different e-commerce platforms (e-commerce
consideration vital information that is provided in written platforms). Our process has reached its initial stage at this
customer reviews. The incorporation of sentiment point. There are a few examples of additional metadata
analysis into recommendation systems is something that that might be added to this data. These cases could be

Authorized licensed use limited to: Zhejiang University. Downloaded on January 08,2025 at 07:44:51 UTC from IEEE Xplore. Restrictions apply.
included. Among the examples are timestamps and on textual customer assessments. This is the next stage
product identification numbers. The inclusion of ratings that needs to be completed. For binary sentiment
and reviews in written form is another option that might categorization, there are supervised learning algorithms
be considered. After the data has been collected, it is put that can be utilized. Support vector machines (SVM),
through a preprocessing step, which involves cleaning it logistic regression, and random forests are a few
up and getting it ready for the analysis of sentiment[10]. examples of the approaches that fall under this category.
There is a possibility that stopwords, punctuation, and The classification of either positive or negative feelings is
special characters will be eliminated during this step of made possible through the use of these procedures. There
the preparation process. In addition, tokens, lowercase is also the possibility of employing techniques for
letters, and special characters may also be eliminated multiclass classification tocategorize the polarity of
from the list. In addition, the process of reducing words to feeling, which can be divided into three distinct
their fundamental forms can be performed through the categories: positive interpretations, negative
utilization of stemming or lemmatization. interpretations, and neutral interpretations. Convolutional
neural networks (CNNs) and recurrent neural networks
Using the phrasing of the evaluations as a basis, the (RNNs) are two examples of deep learning architectures
purpose of this study is to ascertain whether or not the that can be utilized to attain state-of-the-art performance
consumer will suggest the product to others[11]. To be in sentiment analysis tasks and to capture precise
more specific, we are interested in determining whether or correlations that exist within textual data. Both of these
not the consumer appreciates the product to the extent that designs are instances of deep learning. Utilizing deep
they would suggest it to others. Consequently, it is a learning is a method that can be utilized to accomplish
problem of binary classification, which indicates that the both of these objectives.
target variable can either take the value 0 or the value 1
depending on the nature of the situation. To achieve this IV.RESULTS AND DISCUSSION
goal, the following steps are carried out, as shown in the
illustration that can be found below (Fig 1): Using a confusion matrix is one method of summarising
the outcomes of predictions made during a classification
operation. The accurate and inaccurate predictions are
highlighted and categorized into classes. A comparative
analysis is conducted between the predicted and actual
values. The confusion matrices that are present in the
spreadsheet are shown in Figure 13. True positives (TP)
and false positives (FP) are abbreviated as TP and FP,
respectively, and false negatives (FN) and true negatives
(TN), respectively. Where is it located?

Figure 1: Depicts the study'spipeline  TP: The number of consumers (class 1) who have
recommended a product and for whom the predictive
Following the completion of this process, the textual data model has correctly predicted their behavior.
is converted into numerical feature vectors, which may  TN: The proportion of consumers that the predictive
then be utilized as input for machine learning algorithms. model correctly predicted and the proportion of
Immediately following the conclusion of the consumers who did not recommend a product (class
preprocessing stage, this change will take place. Both the 0).
process of feature extraction and the process of feature  FP: The proportion of customers that the predictive
representation can be accomplished through the algorithm identified as class 1 but who did not
application of a variety of different approaches[12]. Bag- propose a product (class 0).
of-words (BoW), term frequency-inverse document  FN: The number of clients that recommended a
frequency (TF-IDF), word embeddings (such as product (class 1) but were categorized as class 0 by
Word2Vec and GloVe), and deep learning-based the prediction system.
representations (such as BERT and ELMo) are some of
the methods that fall under this category. To train To measure the effectiveness of the models that were
machine learning models for sentiment analysis, it is applied or the prediction of the product suggestion that
possible to make use of these feature representations, consumers would make on the test set, a range of
which can extract semantic information from the textual measures were employed, including precision, recall,
input. The usage of these feature representations is what accuracy, and F1-score. They ascertain whether the
makes this a feasible option. prediction algorithms can correctly predict the customer
recommendation, which may be favorable or unfavorable.
The next phase that needs to be completed is to train The information obtained via the use of the confusion
machine learning models to perform sentiment analysis

Authorized licensed use limited to: Zhejiang University. Downloaded on January 08,2025 at 07:44:51 UTC from IEEE Xplore. Restrictions apply.
matrix serves as the basis for each of the four indicators intentions of the reviewer. Thus, another crucial area that
that were previously covered. calls for the use of deep learning and machine learning
techniques is the detection of bogus reviews.
Table 1: Depicts the accuracy, precision, recall, and f1-
scorevaluesofthemodelsused. V.CONCLUSIONS AND FUTURE DIRECTIONS
The goal of this research study is to offer a
comprehensive method for conducting sentiment analysis
in customer evaluations to make product suggestions in
the context of online shopping. In the process of carrying
out the research, the methodology makes use of various
machine-learning techniques. Through the utilization of
advanced machine learning algorithms, we have
The use of customer reviews as a data collection tool for demonstrated the capability to extract sentiment
customer relationship management in online retail is an information from textual consumer comments, categorize
intriguing issue to think about. The accuracy of the sentiments, and incorporate them into product
models mentioned above varied from 80 to 90 percent. recommendation systems. This allows us to provide
The Logistic Regression model performed better than the customers with recommendations that are more
other models in terms of the area under the curve (AUC) personalized and pertinent to their desires and
score and the confusion matrix parameters. It's feasible requirements.
that this model will be developed further to include user-
to-user interactions to gather more data about a particular Several prospective avenues for research and
product or consumer. This would enable the development development in the future are becoming more obvious in
of a very potent tool that would help clients make well- this field as time goes on. These avenues are becoming
informed decisions about their purchases and be an visible in this domain. To begin, the accuracy and
intriguing tool for managing their client relationships. persistence of sentiment categorization models might be
enhanced by conducting additional research into novel
machine learning algorithms and methodologies for
sentiment analysis. This would be a good place to start.
Deep learning architectures and ensemble methods are
two examples of technologies and algorithms that fall
under this category. Furthermore, the examination of the
incorporation of multimodal data sources, which include
photographs, videos, and audio, into sentiment analysis
and recommendation systems may provide more in-depth
insights into the preferences and feelings of customers.
These sources include photos, videos, and audio.

Addressing ethical issues concerning user privacy,

transparency, and algorithmic bias in the context of
sentiment analysis and recommendation systems is
essential to cultivate trust and confidence among users.
Figure 2: Depicts the Confusion matrices of the models This is to foster trust and confidence among users. In the
used. future, research endeavors should give importance to the
building of machine learning models that are both fair and
Further investigation is required to enhance the transparent. The objective of these models should be to
performance measurements even more. Two techniques preserve the privacy of users and to promote equity in
that can be applied to each new application that is created environments that are associated with online commerce.
in line with machine learning principles are opinion
mining and sentiment analysis. It could be challenging to In addition, longitudinal studies are necessary to
spot phony evaluations just by reading them, and there are investigate the impact that sentiment-aware
still many unanswered questions in this area of study. recommendation algorithms have on the level of user
This is true even if sentiment analysis algorithms and satisfaction, engagement, and purchasing behavior in e-
methodologies are producing high-caliber studies and commerce platforms over the course of a longer period.
advancing significantly. Sometimes, phony evaluations Through user surveys and A/B testing trials, researchers
are misinterpreted as real ones and are so heavily can evaluate the efficacy and practicability of sentiment-
manipulated that it is impossible to discern the true based recommendation systems in real-world scenarios.
This includes determining whether or not these systems

Authorized licensed use limited to: Zhejiang University. Downloaded on January 08,2025 at 07:44:51 UTC from IEEE Xplore. Restrictions apply.
are useful. This enables them to identify areas in which [4] G. R et al, "An Effective Copyright Management Principle
using Intelligent Wavelet Transformation based Water marking
there is space for development and work to improve those Scheme," (ACCAI), Chennai, India, 2022, pp. 1-7, doi:
areas. 10.1109/ACCAI53970.2022.9752516.
[5] Sethy, P.K., Panigrahi, M., Vijayakumar, K. et al. Machine
Machine learning has the potential to dramatically learning based classification of EEG signal for detection of
epileptic seizure without snipping. Int J Speech Technol (2021).
increase the relevance, customisation, and efficiency of
https://doi.org/10.1007/s10772-021-09855-7
recommendation systems. In a nutshell, the application of [6] Nuthakki, R., Masanta, P., Yukta, T.N. (2022). A Literature
machine learning to assess customer evaluations for the Survey on Speech Enhancement Based on Deep Neural
goal of producing product recommendations in online Network Technique. In: Kumar, A., Mozar, S. (eds) ICCCE
retail has the potential to make recommendation systems 2021. Lecture Notes in Electrical Engineering, vol 828.
Springer, Singapore. https://doi.org/10.1007/978-981-16-7985-
more efficient. By collecting textual customer feedback, 8_2
e-commerce platforms have the capacity to improve user [7] Zhang, J., Zhao, Y., &LeCun, Y. (2015). Character-level
satisfaction, generate sales, and nurture consumer loyalty Convolutional Networks for Text Classification. Advances in
in the increasingly competitive online marketplace. This Neural Information Processing Systems, 28, 649-657.
[8] Turney, P. D. (2002). Thumbs Up or Thumbs Down? Semantic
is accomplished by gaining an understanding of sentiment Orientation Applied to Unsupervised Classification of Reviews.
patterns and user preferences through the collection of Proceedings of the 40th Annual Meeting on Association for
customer feedback. Continued research and innovation in Computational Linguistics, 417-424. doi:
this field is required in order to reach the full potential of 10.3115/1073083.1073153
[9] Cambria, E., & Hussain, A. (2015). Sentic Computing: A
sentiment-aware recommendation systems and to define Common-Sense-Based Framework for Concept-Level
the future of e-commerce. This is a necessity that cannot Sentiment Analysis. Springer International Publishing. doi:
be avoided. 10.1007/978-3-319-16994-
[10] M. T, et al "IoT based Smart Robotic Design for Identifying
Human Presence in Disaster Environments Using Intelligent
REFERENCES Sensors," (AUTOCOM), Dehradun, India, 2024, pp. 399-403,
doi: 10.1109/AUTOCOM60220.2024.10486106
[1] Kim, Y., & Lee, K. (2020). Sentiment Analysis in E-commerce: [11] A. G, et al, "An Artificial Neural Network Classifier for palm
A Review and Research Agenda. IEEE Access, 8, 135351- Motion categorization based on EMG signal," (ICSES),
135366. doi: 10.1109/ACCESS.2020.3019369 Chennai, India, 2022, pp. 1-11, doi:
[2] S. Rosaline et al, “Predicting Melancholy risk among IT 10.1109/ICSES55317.2022.9914097.
professionals using Modified Deep Learning Neural Network [12] Venkatesan, K., Chandrasekar, A. & Ramesh, P.G.V. On-
(MDLNN),” (CSNT), Apr. 2022, Published, doi: demand DWDM design using machine learning. Soft Comput
10.1109/csnt54456.2022.9787571. 26, 6577–6589 (2022). https://doi.org/10.1007/s00500-022-
[3] Liu, B. (2012). Sentiment Analysis and Opinion Mining. 07181-x.
Synthesis Lectures on Human Language Technologies, 5(1), 1-
167. doi: 10.2200/S00416ED1V01Y201204HLT016

Authorized licensed use limited to: Zhejiang University. Downloaded on January 08,2025 at 07:44:51 UTC from IEEE Xplore. Restrictions apply.