Journal Européen des Systèmes Automatisés

Vol. 57, No. 4, August, 2024, pp. 1145-1153

Journal homepage: http://iieta.org/journals/jesa

Recommendation System for E-Commerce Using Collaborative Filtering

Preeti Patil1 , Sandeep Uddhavrao Kadam2 , E.R. Aruna3 , Amar More4 , Balajee R.M.5 ,
B. Narendra Kumar Rao6*
1
Department of Information Technology, D Y P College of Engineering, Pune 411044, India
2
Department of Computer Engineering, Keystone School of Engineering, Pune 412308, India
3
Department of CSE, Vardhaman College of Engineering, Hyderabad 501218, India
4
School of Computer Engineering, MIT Academy of Engineering, Pune 411011, India
5
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Guntur 522302, India
6
School of Computing, Mohan Babu University, Tirupati 517501, India

Corresponding Author Email: narendrakumarraob@gmail.com

Copyright: ©2024 The authors. This article is published by IIETA and is licensed under the CC BY 4.0 license
(http://creativecommons.org/licenses/by/4.0/).

https://doi.org/10.18280/jesa.570421 ABSTRACT

Received: 6 June 2024 In the swiftly evolving domain of online commerce, the significance of recommendation
Revised: 1 August 2024 systems has risen alongside the rapid expansion of internet usage over the past decade. As
Accepted: 9 August 2024 online commerce continues to thrive, recommendation systems serve a crucial function in
Available online: 27 August 2024 steering users towards pertinent products amidst the expansive online environment.
Nonetheless, conventional collaborative filtering algorithms often encounter challenges
such as sparse data and shifting user preferences, necessitating innovative approaches. Our
Keywords: proposed recommendation system aims to tackle these hurdles by seamlessly blending
recommendation, collaborative filtering, e- collaborative filtering and content-based filtering methodologies. It will offer product
commerce suggestions for both new and existing users. Through thorough examination of users' past
purchasing behaviors, patterns, and feedback, our system customizes recommendations
precisely to cater to existing users' needs. The initial stage involves feature extraction,
wherein both content-based and collaborative features are obtained by creating user
profiles, computing content similarity, identifying related items, generating
recommendations, and suggesting items. Additionally, new users will receive
recommendations for novel and trending products. Leveraging the Walmart Product rating
Dataset, our system continuously enhances recommendations based on evolving user
interactions, thus optimizing engagement and satisfaction levels. This study underscores
the pivotal role of advanced recommendation techniques in transforming the online
commerce landscape, ensuring informed purchasing decisions, heightened user
satisfaction, and increased sales.

1. INTRODUCTION e-commerce space, developing and implementing such

recommendation systems poses a considerable challenge.
In the rapidly evolving landscape of e-commerce, the The development of an effective product recommendation
competition among online platforms is reaching system hinges on the availability and analysis of vast datasets
unprecedented levels. Established giants dominate the market, to gain insights into customer preferences and behaviors.
while emerging companies, often referred to as unicorns, Established e-commerce platforms, with their extensive user
strive to carve out their niche and challenge the dominance of base and transaction history, have access to rich datasets that
major players. However, amidst this intense competition, can be leveraged to train sophisticated recommendation
many new entrants face a critical obstacle: the absence of algorithms. In contrast, new companies, often characterized by
essential sales booster features, particularly robust product limited sales and resources, struggle to accumulate and process
recommendation systems. These systems play a pivotal role in such extensive data. Consequently, they face a daunting task
driving sales, enhancing user experience, and fostering in devising recommendation systems that can rival those of
customer loyalty [1]. their larger counterparts.
In these recent years, significance of product Moreover, the lack of satisfactory open-source solutions
recommendation systems has grown substantially as they offer further exacerbates the challenge for smaller e-commerce
personalized suggestions to users based on their browsing entities [3]. While several recommendation algorithms and
history, purchase behavior, and preferences [2]. These systems frameworks are available in the public domain, they may not
not only facilitate decision-making for customers but also adequately address the unique needs and constraints of
contribute significantly to revenue generation for e-commerce emerging companies [4]. As a result, these companies are left
platforms. However, for smaller companies venturing into the grappling with the dilemma of either investing heavily in

1145
proprietary solutions or compromising on the effectiveness of sentiment analysis can complement recommendation systems
their recommendation systems. by incorporating user sentiment towards products or brands,
Nevertheless, within the realm of AI-ML, a plethora of thereby improving recommendation relevance and user
theories, methodologies, and techniques hold promise for satisfaction.
addressing the pressing need for tailored product Khan et al. [10] conducted a survey on collaborative
recommendation systems. By harnessing the power of these filtering within online recommendation systems. Their study
advanced technologies, new companies can overcome the presents an overview of diverse collaborative filtering
barriers posed by limited resources and data availability to techniques, encompassing user, item, and model-based
develop innovative solutions that cater to their specific approaches, and explores their applications across online
requirements [5]. Through strategic implementation of platforms.
machine learning algorithms, data analytics, and user-centric Other researches also introduced a trust-based collaborative
design principles, emerging e-commerce ventures can not only filtering algorithm tailored for e-commerce recommendation
enhance their competitiveness but also improve user systems. By integrating trust relationships among users, their
satisfaction and drive growth in the fiercely competitive approach significantly boosts recommendation accuracy while
market landscape [6]. effectively tackling the common cold-start problem associated
In this context, this research endeavors to explore the with collaborative filtering. Collaborative Filtering faces the
challenges faced by new e-commerce companies in "cold-start problem" when new users or items are introduced
developing product recommendation systems and to propose to the system. The following strategies can be adapted with
novel approaches and solutions to address these challenges. By New Users like Gather Initial Data using Surveys and
examining existing literature, analyzing case studies, and following some onboarding process, use Hybrid Approaches
leveraging cutting-edge methodologies from the field of like content-Based Filtering. For New Items following can be
machine learning, this study seeks to provide insights and observed content-Based Filtering involving item features,
guidance for emerging companies seeking to enhance their Hybrid Approaches and exposure to promotional strategies.
recommendation capabilities and gain a competitive edge in Wang and Wang [11] developed recommendation system
the e-commerce arena. for e-commerce was developed, leveraging an enhanced
The unique challenges in developing recommendation collaborative filtering algorithm. Their approach introduces
systems for new and smaller e-commerce companies includes enhancements to traditional collaborative filtering methods to
Cold Start Problems, Data Sparsity, Scalability, Resource overcome limitations such as sparsity and scalability, thereby
Constraints, during Real-time Processing and limited improving recommendation quality.
infrastructure. Few of the AI&ML techniques used to address Anwar and Uma conducted a comparative study was
these challenges are Hybrid Recommendation Systems, conducted to examine various recommendation system
Matrix Factorization, Incremental Learning, Graph-based approaches, with a focus on movie and multimedia
Models and Deep Learning models. recommendation utilizing collaborative filtering [12]. The
Overall, the integration of effective product analysis delves into the strengths and weaknesses of different
recommendation systems is crucial for the success and collaborative filtering techniques, emphasizing the
sustainability of e-commerce ventures, particularly for newer significance of algorithm selection in the design of
entrants aiming to establish themselves in the highly recommendation systems.
competitive market. Through innovative strategies and Priya and Mansoor Hussain [13] have discussed three
leveraging advancements in technology, these companies can important techniques of creating a recommendation system-
leverage recommendation systems to drive sales, foster Collaborative Filtering, Content based and Hybrid. They have
customer engagement, and propel their growth trajectory [7]. also discussed the existing e-commerce websites and the type
of Recommendation they use.
Overall, the literature reviewed underscores the importance
2. LITERATURE SURVEY of collaborative filtering and related techniques [8] in
developing effective product recommendation systems for e-
Product recommendation systems have become integral commerce. By leveraging user-item interactions, textual data,
tools for enhancing user experience and driving sales in the sentiment analysis, and trust relationships, researchers aim to
competitive landscape of e-commerce. Leveraging various enhance recommendation accuracy, address scalability
methodologies and techniques, researchers have explored challenges, and improve user satisfaction in the competitive e-
innovative approaches to developing effective commerce landscape.
recommendation systems tailored to the specific needs of
online platforms. The objective of this literature review is to
offer an overview of recent research concerning product 3. PROPOSED METHODOLOGY
recommendation systems, with a particular emphasis on
collaborative filtering and associated methodologies. 3.1 Proposed work
Khatter et al. [8] proposed an e-commerce product
recommendation system employing collaborative filtering and The proposed work aims to develop a modern
textual clustering techniques. Their approach combines recommendation system for e-commerce platforms, centered
collaborative filtering, which analyzes user-item interactions, around collaborative filtering techniques. By leveraging users'
with textual clustering to enhance recommendation accuracy purchase history and search behavior, the system will analyze
by considering product descriptions and user preferences. patterns and preferences to offer personalized product
Dubey et al. [9] introduced an expanded opinion lexicon suggestions. Additionally, content-based filtering will be
combined with machine learning-based sentiment analysis of integrated to ensure diverse recommendations tailored to
tweets. Although not directly related to collaborative filtering, individual tastes. User interactions with the cart will serve as

1146
feedback loops for continuous refinement of recommendations, sharing diverse datasets. The Walmart Product rating Dataset
enhancing overall user experience and engagement. Through collection process involves gathering comprehensive
this research, we seek to demonstrate the potency of information about products added to users shopping carts on
collaborative filtering in delivering highly personalized an e-commerce platform. The data set consists of 5000 rows
recommendations, ultimately driving increased sales and user and 31 columns. Out of which Product Id, Product Name,
satisfaction. By setting a new standard in e-commerce Product Category, Product Price, Item Number, Rating,
personalization, this system aims to improve customer loyalty, Review. Each product typically has attributes such as Product
retention, and overall business growth in the competitive Id, Product Name, Product Category, Item Number, Rating,
online marketplace. A Collaborative filtering approach is Review and Price.It includes User review and User ratings.
essential for e-commerce recommendation systems because of The data entered in the Users rating column are number format
its ability to provide personalized recommendations by only supports the values With values ranging from 1 to 5. The
leveraging user behavior patterns and preferences [9]. Its dataset contains different category of products some of them
domain-independence nature makes it more versatile across are Beauty, Health, Household essentials, Personal care,
products without requiring specific content information. Sports and Outdoors, etc,
By using the “marketing_sample_for_walmart_com-
3.2 System architecture walmart_com_product_review” Dataset, e-commerce
platforms gain insights into user preferences, purchasing
The system architecture comprises three main components: behaviour, and product popularity. This dataset serves as the
data processing, recommendations, and final outcome foundation for developing and optimizing recommendation
generation as in Figure 1. The data processing module receives systems, enhancing user experience, and driving sales by
and preprocesses cart product data sets, extracting relevant providing personalized and relevant product suggestions.
features such as product attributes and user interactions. These Preprocessing:
processed data are then fed into two recommendation modules: The data processing workflow involves many steps to
content-based & collaborative filtering. prepare the dataset for further analysis as well as for modeling.
This content-based recommendation module utilizes Initially, the pandas dataframe is utilized to load and organize
product attributes to suggest items similar to those in cart, the data. During data preprocessing, if our dataset contains
ensuring diversity and relevance. Meanwhile, the missing values in the rating field, we fill them with zeros.
collaborative filtering [10] method analyzes user behavior and Conversely, if crucial fields such as product name and details
interactions to generate personalized recommendations based are absent, we exclude the corresponding rows to uphold data
on similarities between users & items. integrity. Outliers are addressed by either removing them if
Finally, recommendations from both modules are combined they are extreme or transforming them using techniques like
and further refined based on user feedback and interactions robust scaling to mitigate their impact. To eliminate irrelevant
with the cart. The final outcome is a set of highly personalized rows and columns, Singular Value Decomposition (SVD) can
product recommendations tailored to individual user be utilized. Data splitting entails dividing the dataset into
preferences. This architecture enables the system to deliver training, validation, and testing sets to assess machine learning
accurate and diverse suggestions, enhancing user satisfaction models.
and driving increased sales for e-commerce platforms. An The training data is utilized for model training, the
approach like the below can be used where α is a weight factor validation data for hyperparameter tuning and performance
between 0 and 1, i is the item, and u is the user. assessment, and the testing data for evaluating the final
model's performance on unseen data. These preprocessing
𝐹𝑖𝑛𝑎𝑙 𝑆𝑐𝑜𝑟𝑒(𝑖, 𝑢) − 𝛼 × 𝐶𝐵 𝑆𝑐𝑜𝑟𝑒(𝑖, 𝑢) + (1 − 𝛼) × 𝐶𝐹 𝑆𝑐𝑜𝑟𝑒(𝑖, 𝑢) steps ensure the data is cleansed, standardized, and suitable for
analysis or modeling, thereby yielding more precise and
dependable results. Additionally, label processing is
performed to encode categorical variables or convert them into
numerical representations suitable for machine learning
algorithms. This may involve techniques such as one-hot
encoding or label encoding. Few limitations with
marketing_sample_for_walmart_com-
walmart_com_product_ review dataset includes Sampling
Bias, Data Quality Issues, Temporal Dynamics, Contextual
Limitations, Sentiment Analysis Challenges, Cultural and
Demographic Factors.
To ensure the proposed model attains high accuracy, several
strategies are indispensable. These strategies encompass
addressing missing values by substitution with 0, managing
imbalanced data, determining an optimal value for K via cross-
validation for the K-nearest neighbors (KNN) algorithm,
Figure 1. Proposed architecture selecting pertinent features and scaling them suitably,
choosing an appropriate distance metric, evaluating
3.3 Database description generalization performance through cross-validation, and
implementing dimensionality reduction using Singular Value
Samples from “marketing_sample_for_walmart_com- Decomposition (SVD). By incorporating these strategies,
walmart_com_product_review” dataset is taken from the optimizing parameters, and refining preprocessing steps, the
Kaggle website, which serves as a vital hub for accessing and model can achieve high accuracy.

1147
 Collaborative filtering hinges on user-item interactions, personalized suggestions aligned with their preferences and
whereas content-based filtering leverages item attributes. past interactions. Through Content-Based Filtering, users
Integrating both methodologies amalgamates their merits to receive tailored recommendations that enhance their shopping
augment recommendation precision. The optimization process experience and increase the likelihood of finding products of
entails several steps. If our dataset contains missing values in interest. Content-Based Filtering techniques include Feature
the rating field, we replace them with zeros. Conversely, if Extraction and User Profiling. The algorithms associated are
vital fields like product name and details are absent, we TF-IDF (Term Frequency-Inverse Document Frequency),
eliminate the corresponding rows to uphold data integrity. Cosine Similarity, k-Nearest Neighbors (k-NN), Naive Bayes,
Outliers are addressed by either removing extreme cases or Decision Trees, Deep Learning (e.g., CNNs for images, RNNs
employing techniques like robust scaling to diminish their for text).
impact. Additionally, irrelevant rows and columns are filtered
out using Singular Value Decomposition (SVD). Another 3.6 Collaborative filtering method
aspect of optimization involves tuning hyperparameters.
During this process, you experiment with settings such as the This is a recommendation method that suggests products
number of neighbors in collaborative filtering or the similarity based on user behavior and preferences, leveraging similarities
threshold in content-based filtering. Techniques like grid among users or items.
search or random search aid in identifying the optimal To achieve collaborative filtering, data is first imported
hyperparameters for enhanced model performance. These from a dataset containing items, such as ratings and purchases.
meticulous steps ensure that your recommendation system is After preprocessing to address missing values and outliers, the
finely calibrated to provide accurate and pertinent product similarity between users or items is computed. This involves
suggestions to users. constructing a utility matrix that represents user-item
preferences, where most values are initially unknown and set
3.4 Recommendation to 0. Subsequently, the utility matrix is transposed. Singular
Value Decomposition (SVD), a dimensionality reduction
The recommendation process comprises two main methods: technique, is then applied to decompose the matrix into three
content-Based Filtering & collaborative Filtering. Product components: the original matrix (A), an orthogonal left
recommendations in content-based filtering are generated singular matrix (U), and a diagonal matrix (S). Following this,
through evaluating the similarity of items in the dataset. This a correlation matrix is generated by computing correlations
method utilizes features such as product attributes, between items based on the ratings given by users who have
descriptions, and user preferences to identify items that closely purchased the same products. Finally, the recommendation
match those already in the user's cart. Recommendations are system presents the top 10 products to the customer based on
made by selecting the items with highest similarity scores to the purchase history of other customers on the website. Finally,
the ones already present. the recommended products are displayed to the user, providing
On the other hand, Collaborative Filtering recommends personalized suggestions that align with their preferences as
products by analyzing user-item interactions across the entire well as past interactions. Collaborative Filtering enhances user
dataset. This method identifies patterns and similarities in user experience by delivering tailored recommendations and
behavior, such as purchase history or item ratings, to generate fostering engagement in e-commerce platforms. Collaborative
personalized recommendations for each user. By leveraging Filtering techniques includes User-Item Interactions and
the collective wisdom of all users, Collaborative Filtering can Matrix Factorization [14]. The algorithms associated are User-
suggest products that are popular among similar users or items Based Collaborative Filtering, Item-Based Collaborative
frequently purchased together. Filtering, Matrix Factorization, Singular Value
By employing both Content-Based & Collaborative Decomposition (SVD) and Neural Collaborative Filtering.
Filtering methods, the recommendation process ensures a The collaborative filtering recommends products by finding
comprehensive and personalized approach in suggesting similar products to the ones the user has interacted with and
products that align with user preferences & interests. suggesting those similar products.

3.5 Content-based filtering 3.7 Equations

This recommendation method emphasizes suggesting 𝛴𝑗𝐶𝑁(𝐼) sin(𝐼 ⋅ 𝑗) ⋅ 𝑟(𝑈𝑗)

𝑟(𝑢𝑖 ) = (1)
products that closely resemble those with which a user has 𝛴𝑗 𝐶𝑁(𝐼)⃓⃓ sin(𝐼 ⋅ 𝑗)
previously interacted. The process begins by appending the list
of products and orders, forming a dataset that captures user- Eq. (1) is the predicted rating for user U on item i, where
item interactions. Subsequently, the data is transformed into N(I) is the neighborhood of similar items to item I, sin(I, j) is
Compressed Sparse Row (CSR) format, facilitating efficient the similarity between items I and j, r(uj) is the actual rating of
computation. We are going to perfrom TF – IDF vectorization. user U on item j.
The value obtained by perfroming multiplication of TF score
and IDF score is assign to each product in the data set.A model 𝑛(𝑑) (𝑡)
𝑇𝐹(𝑡, 𝑑) = (2)
is then created using the Nearest Neighbors algorithm, which 𝑛(𝑑)
identifies products similar to those that user has engaged with,
in past. The model is fitted using the CSR matrix, enabling it Eq. (2) is the term frequency of the word in the product
to efficiently process large datasets. Recommendations are description field, where n(d)(t) represents number of times a
formulated by analyzing a user's previous orders, and the word appears in the product description filed, n(d) represents
suggested products are then added to the result. Finally, the total number of times the word or term occurs in remaining
recommended products are displayed to the user, providing product description filed.

1148
 𝑛(𝑑) comprising Collaborative Filtering and Content-Based
𝐼𝐷𝐹(𝑡, 𝑑) = 𝑙𝑜𝑔 (( ) + 1) (3) Filtering, attains a 97% accuracy rate, surpassing other
𝑛(𝑑) (𝑡)
methods [17] such as Linear Support Vector Machine (LSVM),
Decision Tree Algorithm (DT), Linear Regression (LR), and
Eq. (3) is the Inverse Document Frequency. Where t Collaborative Filtering and Textual Clustering (CB and TC).
represents term or word in the description field, n(d) represents The accuracy of various algorithms is presented in Table 1,
total number of description fields and n(d)(t) represents the and it's visually represented through a bar graph. Our proposed
number of description fields containing the term and D model, Collaborative Filtering and Content-Based Filtering
represents entire collection of description fields. [18] (Cf and CBF), achieves an accuracy of 97%. In
comparison, Linear Support Vector Machine (LSVM)
𝑇𝐹𝐼𝐷𝐹(𝑡, 𝑑, 𝐷) = 𝑇𝐹(𝑡, 𝑑) × 𝐼𝐷𝐹(𝑡, 𝐷) (4) achieves 96% accuracy, Decision Tree (DT) reaches 94.54%,
Linear Regression (LR) scores 94%, and Collaborative
Eq. (4) is the TFIDF score, where t represents a term (word) Filtering and Textual Clustering (CF and TC) achieves 93%.
in a document, d represents a specific document [15], D By comparing these models, it's evident that our proposed
represents the entire collection of documents. model achieves the highest accuracy [19] as in Figure 2.
Algorithm in pseudo format Table 1. Performance metrics
1. Import necessary libraries: “numpy”, “pandas”,
matplotlib.pyplot”, “sklearn”, “TruncatedSVD from Accuracy F-
sklearn.decomposition”. Model Precision Recall
(%) Measure
2. Load the dataset from the specified file path CF and
97 94 95 95
3. (`Walmart_product_rating.csv`). CBF
4. Transpose the ratings utility matrix. LSVM 96 93 95 94
5. Perform Truncated SVD with 10 components on the DT 94.54 95 95 95
transposed matrix. LR 94 93 95 94
CB and TC 93 91 95 93
6. Compute the correlation matrix from the decomposed
matrix.
7. Define a product ID 'i'. Retrieve the index of 'i' from
the product names and compute the correlation of 'i'
with other products.
8. Recommend products with a correlation coefficient
greater than 0.90, excluding the already bought
product 'i'.
9. Vectorize the product descriptions using TF-IDF
vectorizer.
10. Fit K-Means clustering algorithm with 10 clusters to
the vectorized data.
11. Plot the clusters and print the top terms per cluster.

Final outcome
The final outcome of the recommendation system is a user-
friendly interface that seamlessly integrates both Content- Figure 2. Bar graph showing accuracy of various algorithms
Based and Collaborative Filtering methods. This interface
provides personalized product recommendations tailored to
each user's preferences and interactions, enhancing the overall
e-commerce experience. By leveraging Content-Based
Filtering, users receive suggestions based on the similarity of
products they have interacted with previously. Additionally,
Collaborative Filtering [16] takes into account similarities
among users or items to offer recommendations that align with
individual preferences. The integration of these methods
ensures that users receive diverse and relevant product
suggestions, increasing the likelihood of finding items of
interest and improving user satisfaction. Through the user-
friendly interface, customers can easily navigate and explore
personalized recommendations, leading to enhanced
engagement and increased sales for the e-commerce platform.

Figure 3. Accuracy, precision, recall, and F-measure are

4. RESULTS examples of performance metrics

Table 1 presents performance metrics for different Figures 2 and 3 illustrates performance metrics of various
algorithms, contrasting them with the proposed algorithms algorithms that are shown in Table 1. This representation
outlined. Table 1 offers concise insights into the performance provides clear information about the performance of various
of diverse machine learning algorithms. Our proposed model, machine learning algorithms. Our proposed model achieves

1149
higher accuracy than compare to above algorithms. The Screen
shots of implementation are provided in Figures 4-10.

Figure 9. Product recommendation

Figure 4. Dash board

Figure 10. Output

The user-item rating matrix in Table 2 shows the ratings

Figure 5. About page given by the users to the items. A blank cell means that the
user has not rated the item.
Figure 11 illustrates which items are popular among the
users and which items are missing ratings

Table 2. The user-item rating matrix

Items
Users
Item-1 Item-2 Item-3 Item-4 Item-5
User-1 4 3 5
User-2 4 3 5
User-3 5 4 3
Figure 6. Create an account page User-4 5 4 3
User-5 3 5 4
User-1 4 3 5

Figure 7. Login into account page

Figure 11. User-item rating matrix

The heat map shows the ratings as colors on a grid. A heat

map can help you see which items are popular among the users
and which items are missing ratings.
Figure 8. Welcome to dashboard page The user profile matrix in Table 3 shows the features or the

1150
embedding of the users based on their ratings. The features are This correlation matrix in Table 6 shows the correlation
represented as vectors of numbers, which capture the semantic between each pair of items based on their decomposition into
and contextual relationships between the users and the items. the three components.
Figure 12 illustrates the features or the embedding of the It represents providing the recommendations to the product
users based on their ratings. searched by the user whose correlation values is equal to 0.9.
From Table 7, recommended products with correlation
Table 3. The user-profile matrix coefficient greater than 0.90 (excluding 'Item 1' i.e. our
searched product is item 1) is Item2.
S.No
Features TF-IDF vectorization converts text data such into the
Feature1 Feature2 Feature3 Feature4 Feature5 numerical vectors as in Table 8 by assigning weights to words
1 0.41 0.36 0.22 0.72 0.19 based on their frequency and importance, facilitating effective
2 0.37 0.39 0.24 0.68 0.22
3 0.43 0.34 0.21 0.74 0.18
analysis of textual data such as product descriptions. Since we
4 0.36 0.41 0.23 0.66 0.24 have 3 products, each represented by TF-IDF vectors.
5 0.39 0.38 0.25 0.70 0.21
Table 6. Correlation matrix

S. Components
Items
No Component-1 Component-2 Component-3
1 Item-1 1 0.985 0.029
2 Item-2 0.985 1 0.209
3 Item-3 0.209 0.209 1

Table 7. Item-correlation table

Items
S.No
Item-1 Item-2 Item-3
1 1 0 0
Figure 12. User-profile matrix 2 0 0.985 0
3 0 0 0.209
The user-profile matrix is represented by bar chart, which
show the features or the embedding of the users and the items Table 8. Term frequency inverse document frequency
as bars of different heights. A bar chart can help you compare vectorization
the similarities and differences between the users and the items
based on their features. Items
The item profile matrix in Table 4 shows the features or the Item-1 Item-2 Item-3
embedding of the items based on their ratings. The features are [0.0,
[0.6191303, [0.0, 0.0,
0.6191303,
represented as vectors of numbers, which capture the semantic Associated 0.0, 0.78528828,
0.0,
and contextual relationships between the items and the users. Vectors 0.78528828, 0.0,
0.78528828,
0.0, 0.0] 0.6191303]
0.0]
Table 4. The item-profile matrix

S.No
Features 5. CONCLUSION
Feature1 Feature2 Feature3 Feature4 Feature5
1 0.32 0.45 0.12 0.67 0.23
In conclusion, this work has successfully developed an
2 0.28 0.51 0.16 0.63 0.27
3 0.35 0.42 0.18 0.69 0.21 efficient e-commerce product recommendation system,
4 0.26 0.53 0.14 0.61 0.29 addressing the increasing demand for personalized
5 0.31 0.47 0.17 0.65 0.25 suggestions in today's competitive online market. By
integrating Content-Based and Collaborative Filtering
These are the results of performing Truncated SVD with 3 techniques, our system ensures extensive recommendation
components on the transposed ratings utility matrix. Each row coverage, accommodating diverse user preferences. Through
represents a user, and each column in Table 5 represents a meticulous exploration and preprocessing of the
component extracted by the Truncated SVD algorithm. This “marketing_sample_for_walmart_com_walmart_com_produ
process helps in reducing the dimensionality of the data while ct_review” Dataset, coupled with the deployment of
retaining its most important features. recommendation models, our system achieves optimal
performance in delivering relevant product suggestions.
Table 5. Truncated singular value decomposition matrix Collaborative filtering accurately identifies user patterns and
similarities, while content-based filtering enriches
Components recommendation variety. The user-friendly interface further
S.No Users
Component-1 Component-2 Component-3 underscores the system's efficacy, providing users with a
1 0.32 4.120 2.091 -0.040 smooth shopping experience. Looking ahead, future research
2 0.28 3.664 -0.161 -0.199 could explore integrating additional data sources and advanced
3 0.35 4.946 -0.099 0.114 machine learning techniques to enhance the system's
4 0.26 4.385 -0.147 0.163 adaptability in meeting evolving user needs within the
5 0.31 4.203 2.664 -0.038
dynamic e-commerce landscape, ultimately driving increased

1151
sales and customer retention. https://doi.org/10.1016/j.knosys.2013.03.012
Future Enhancements [7] Ekstrand, M.D., Riedl, J.T., Konstan, J.A. (2011).
The future prospects for e-commerce product Collaborative filtering recommender systems.
recommendations utilizing collaborative filtering and content- Foundations and Trends® in Human-Computer
based filtering encompass several avenues for advancement Interaction, 4(2): 81-173. https://doi.org/10.1007/978-3-
tailored to the project's title. Firstly, enhancing the 540-72079-9_9
recommendation system by integrating hybrid techniques that [8] Khatter, H., Arif, S., Singh, U., Mathur, S., Jain, S.
merge collaborative filtering with content-based filtering can (2021). Product recommendation system for E-
notably enhance recommendation accuracy and diversity, commerce using collaborative filtering and textual
aligning closely with the project's aim of leveraging both clustering. In 2021 Third International Conference on
methods for optimal outcomes. Additionally, exploring Inventive Research in Computing Applications
contextual information such as user location, time of day, or (ICIRCA), pp. 612-618.
device type can further enrich the relevance of https://doi.org/10.1109/ICIRCA51532.2021.9544753
recommendations, enhancing the personalized shopping [9] Dubey, G., Kumar, S., Kumar, S., Navaney, P. (2020).
experience as envisioned in our project. Extended opinion lexicon and ML-based sentiment
Moreover, incorporating real-time updates and dynamic analysis of tweets: A novel approach towards accurate
user feedback mechanisms into the recommendation system classifier. International Journal of Computational Vision
will ensure that suggestions remain current and reflective of and Robotics, 10(6): 505-521.
evolving user preferences, crucial for maintaining user https://doi.org/10.1504/IJCVR.2020.110640
engagement and satisfaction in the competitive e-commerce [10] Khan, B.M., Mansha, A., Khan, F.H., Bashir, S. (2017).
landscape. Embracing emerging technologies like deep Collaborative filtering based online recommendation
learning (DL) and natural language processing (NLP) holds systems: A survey. In 2017 International Conference on
vast potential for extracting new insights from user behavior Information and Communication Technologies (ICICT),
data, complementing the collaborative filtering approach and pp. 125-130.
enhancing the system's capabilities in providing tailored https://doi.org/10.1109/ICICT.2017.8320176
recommendations. Furthermore, expanding collaborative [11] Wang, X., Wang, C. (2017). Recommendation system of
filtering beyond traditional user-item interactions to include e-commerce based on improved collaborative filtering
social networks or influencer recommendations can deepen the algorithm. In 2017 8th IEEE International Conference on
recommendation process, offering users a more Software Engineering and Service Science (ICSESS), pp.
comprehensive and personalized shopping journey. Lastly, 332-335. https://doi.org/10.1109/ICSESS.2017.8342926
ongoing research and experimentation with innovative [12] Ricci, F., Rokach, L., Shapira, B. (2011). Introduction to
algorithms and data processing techniques will be vital for our recommender systems handbook. In Recommender
project to stay ahead in recommendation system innovation Systems Handbook, pp. 1-35. Springer, Boston, MA.
and effectively meet the evolving demands of the e-commerce [13] Priya, S., Mansoor Hussain, D. (2017). Recommendation
market. systems for E commerce: A review. International Journal
of Advanced Research in Computer and Communication
Engineering, 6(4): 500-504.
REFERENCES https://doi.org/10.17148/IJARCCE.2017.6496
[14] Koren, Y. (2008). Factorization meets the neighborhood:
[1] Linden, G., Smith, B., York, J. (2003). Amazon.com A multifaceted collaborative filtering model. In
recommendations: Item-to-item collaborative filtering. Proceedings of the 14th ACM SIGKDD International
IEEE Internet Computing, 7(1): 76-80. Conference on Knowledge Discovery and Data Mining,
https://doi.org/10.1109/MIC.2003.1167344 pp. 426-434. https://doi.org/10.1145/1401890.1401944
[2] Konstan, J.A., Riedl, J. (2012). Recommender systems: [15] Dev, A.V., Mohan, A. (2016). Recommendation system
From algorithms to user experience. User Modeling and for big data applications based on set similarity of user
User-Adapted Interaction, 22(1-2): 101-123. preferences. In 2016 International Conference on Next
https://doi.org/10.1007/s11257-011-9112-x Generation Intelligent Systems (ICNGIS), Kottayam,
[3] Anwar, T., Uma, V. (2021). Comparative study of India, pp. 1-6.
recommender system approaches and movie https://doi.org/10.1109/ICNGIS.2016.7854058
recommendation using collaborative filtering. [16] Kumar, M., Yadav, D.K., Singh, A., Gupta, V.K. (2015).
International Journal of System Assurance Engineering A movie recommender system: Movrec. International
and Management, 12: 426-436. Journal of Computer Applications, 124(3): pp. 7-11.
https://doi.org/10.1007/s13198-021-01087-x [17] Walunj, S.G., Sadafale, K. (2013). An online
[4] Park, Y.J., Tuzhilin, A., Han, I. (2008). An empirical recommendation system for e-commerce based on
comparison of approaches for combining multiple apache mahout framework. In Proceedings of the 2013
recommender systems. Information Fusion, 9(3): 345- annual conference on Computers and people research, pp.
360. 153-158. https://doi.org/10.1145/2487294.2487328
[5] Lops, P., Gemmis, M., Semeraro, G. (2011). Content- [18] Chen, H.W., Wu, Y.L., Hor, M.K., Tang, C.Y. (2017).
based recommender systems: State of the art and trends. Fully content-based movie recommender system with
In Recommender Systems Handbook, pp. 73-105. feature extraction using neural network. In 2017
https://doi.org/10.1007/978-0-387-85820-3_3 International Conference on Machine Learning and
[6] Bobadilla, J., Ortega, F., Hernando, A., Gutiérrez, A. Cybernetics (ICMLC), pp. 504-509.
(2013). Recommender systems survey. Knowledge- https://doi.org/10.1109/ICMLC.2017.8108968
Based Systems, 46: 109-132. [19] Zhao, Z.D., Shang, M.S. (2010). User-based

1152
collaborative-filtering recommendation algorithms on Knowledge Discovery and Data Mining, pp. 478-481.
hadoop. In 2010 Third International Conference on https://doi.org/10.1109/WKDD.2010.54

1153