See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/359039399

E-commerce platform based on Machine Learning Recommendation System

Conference Paper · November 2021

DOI: 10.1109/IMTIC53841.2021.9719822

CITATIONS READS
15 3,970

3 authors:

Muhammad Tahir Rabia Enam

Sir Syed University of Engineering and Technology Sir Syed University of Engineering and Technology
43 PUBLICATIONS 227 CITATIONS 37 PUBLICATIONS 275 CITATIONS

SEE PROFILE SEE PROFILE

Syed Muhammad Nabeel Mustafa

NED University of Engineering and Technology, Karachi
28 PUBLICATIONS 118 CITATIONS

SEE PROFILE

All content following this page was uploaded by Muhammad Tahir on 18 May 2022.

The user has requested enhancement of the downloaded file.

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/359039399

E-commerce platform based on Machine Learning Recommendation System

Conference Paper · November 2021

DOI: 10.1109/IMTIC53841.2021.9719822

CITATIONS READS
0 30

3 authors:

Muhammad Tahir Rabia Enam

Sir Syed University of Engineering and Technology Sir Syed University of Engineering and Technology
22 PUBLICATIONS 33 CITATIONS 29 PUBLICATIONS 119 CITATIONS

SEE PROFILE SEE PROFILE

Syed Muhammad nabeel Mustafa

NED University of Engineering and Technology, Karachi
8 PUBLICATIONS 1 CITATION

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Process model for Virtual reality View project

Improvising Security and Privacy Vulnerabilities in Smart Health View project

All content following this page was uploaded by Syed Muhammad nabeel Mustafa on 09 March 2022.

The user has requested enhancement of the downloaded file.

 E-commerce platform based on Machine Learning
Recommendation System
Muhammad Tahir Rabia Noor Enam Syed Muhammad Nabeel Mustafa
Department of Software Engineering Department of Computer Engineering Dept. of Computer Science & I.T
Sir Syed Uni. of Eng. and Tech, Khi Sir Syed Uni. of Eng. and Tech, Khi NED Uni. of Eng. and Tech, Khi
[email protected] [email protected] [email protected]

Abstract—Information overload is one of the potential setbacks the customer and so improving total profit, which encompasses
to many e-commerce platform users. It is very important to filter the functional goals of recommendation systems including
the media and the choices that are overwhelming for internet relevancy, serendipity and diversity [6]. A recommendation
users while making buying decisions using online stores. To
solve this problem, recommendation systems are used widely. system can help clients quickly find a wide range of things
A recommender system helps users find a product of their own that they are interested in. The popularity of this effective
choice by filtering and prioritizing and effectively generating the suggestion system is growing by the day because it is simple
relevant information to its users. The purpose of a recommender and reliable for a client to purchase online and find the best
system is to save time and hassle of searching through the selections for them.
World Wide Web, instead it generates specific and relevant
content that promotes online transaction and bring satisfaction The Recommender system aims to provide users with product,
to the users of e-commerce platforms. The proposed system is service and information recommendations based on their in-
an e-commerce platform based on an apparel recommendation terests, taking into account their needs and preferences. A user
system that recommends products on the foundation of the user’s would undoubtedly prefer a website that suggests something
preferences. beneficial to him over one that forces him to browse through
Index Terms—Apparel recommender system, Content-based
apparel recommendation, Machine Learning, User preferences
the site to find the products they require. A recommender sys-
tem’s fundamental function is to predict a user’s preferences
by comparing them to those of another group of users [6]. The
recommendation systems are used in different aspects of uses
I. I NTRODUCTION
for example what clothes to buy, types of friends to make,
In recent years, e-commerce has been discovered to have or what kind of online news content to consume. They also
a strong link to consumer pleasure, and success is always provide suggestions from the data extracted from a given user
founded on customer trust. As the use of the internet for shop- profile or the ratings given for an item. Artificial intelligence,
ping became more prevalent, customers are confronted with machine learning, deep learning, and computer vision are
the difficult task of sifting through a large number of product among the primary technologies supporting the development
possibilities to discover the one they require [1]. Artificial of intelligent garment recommendation systems and smart
intelligence (AI), specifically computational intelligence and shopping gadgets. Apparel suggestions serve a unique purpose
machine learning methods and algorithms, have been used to that they not only promote similar products to match users’
increase prediction accuracy and tackle data sparsity and cold current dressing styles, but also provide personalized styling
start difficulties in the creation of recommender systems [2]. In tips to help users gain a better understanding of personalized
the field of e-commerce, the impact of machine learning and styling [7].
deep learning is expanding [3]. These domains’ algorithms Smart apparel recommendation systems, inspired by tradi-
aid in increasing sales and optimizing different parts of e- tional styling services aim to recommend appropriate apparel
commerce operations, from product selection to effective prod- to specific people based on concepts obtained from design
uct ordering. Recommendation system has recently attracted knowledge and professionals’ expertise with computer intelli-
a lot of interest and is being used in a variety of industries gence technology [8]. The recommendation system of apparel
[4]. The exponential growth of the e-commerce business has depends on the suggestion which was carried out manually. To
highlighted the need for effective recommendation systems in help the customer to get the product with customization they
today’s world [5]. Recommendation systems are data-filtering wished for, the system puts on some good recommendations
systems that attempt to anticipate a user’s preferences for one for the customer to increase the satisfaction of customer and
thing over another. Movies, books, research articles, search salesperson gets benefit with more of his products sold [9].
queries, social tags, products, financial services, restaurants, This paper proposes an e-commerce platform based on an
jobs, universities, friends, and other applications all use recom- apparel recommendation system. The application collects the
mendation algorithms. The basic purpose of a recommendation data from the users and then constructs an apparel recommen-
system is to boost product sales by providing a relevant item to dation system using machine learning algorithms, as shown in
 tively segment clients who purchased clothes products. The
primary goal of this research is to identify the various probable
associations of customers in terms of brand, product, and price
based on their purchasing behaviours. The results show that
the clusters created by the algorithm based on K-Means are
similar, and that the results are acceptable based on feedback
from existing customers, and that the algorithm satisfies the
customers’ requirements based on the amount of money (price
range) they want to spend while doing online shopping. Unlike
traditional systems that rely on a user’s previous purchases and
history, this project in [17] tries to generate recommendations
based on an image of a product provided by the user, because
many times individuals see something they like and look for
things that are similar to it.
Fig. 1. System Diagram of the Recommender System
III. M ETHODOLOGY
In this paper, we aim to put forward a method to construct
Fig 1. above. The recommendation system provides users with an apparel recommendation system. We have defined the
quality things, saving them time. modeling and construction phases of an e-commerce apparel
recommendation system. The prediction will be made based
II. L ITERATURE S URVEY on content-based filtering. The most popular sort of filtering
The authors in [10] developed an improved collaborative system is content-based filtering. These systems rely on a
filtering recommendation algorithm and recommendation strat- user’s rating while constructing a profile to obtain preliminary
egy. naive Bayes and K-means played a pivotal role in the information about a user in order to avoid not knowing a
creation on this algorithm. The results proved it an efficient new user. Two sorts of information are primarily emphasized
recommender system. The system proposed in [11] comprises while creating a user profile: the user’s preferences and
of an interactive environment in which a user interacts with the user’s engagement with the recommendation system.
many modules that perform tasks such as data collection It merely suggests items based on a comparison of the
from internet sources, knowledge extraction, clustering, and item’s content and the user’s profile. These systems’ engines
trend/product suggestion. The Clustering Module integrates compare a user’s positively rated item with an item that
multiple grouping algorithms and provides a consensus for user has not yet rated. Users will be recommended the
data clustering. At the same time, the Product Recommender things that have the most similarities. As shown in Fig 2.
and Feedback module collects feedback from designers on The user B and user D both gave ratings to the cell phone
various fashion items and suggests more relevant goods based and cell phone case. Since user A also likes the cell phone,
on their preferences. the system will recommend to the user, a cell phone case also.
[12] performed experiments on the selected trusted data to
make the best recommendations. The similarity was calculated
between the users so that better recommendations can be
made. Then, this similarity is added to the weight factor of the
improved slope one algorithm. The authors of [13] investigated
the k-means-singular value decomposition (k-means-SVD) for
dimensionality reduction and clustering approaches based on
the algorithm of created collaborative filtering recommenda-
tion. In [14], the suggested system will be based on user
evaluations, and it will determine the sentiment score, the most
important features, and their associated values for each item
using natural language processing algorithms. The text mining
results will be utilized to inform the recommendation process.
In [15], Advanced data-analysis techniques were used to
analyze the product data sets in this article, and the results
were positive. The product quality has been accelerated by
extracting product features, product quality, and several other
components of the user evaluation. Based on customer reviews,
a product recommendation has been made to assist the cus- Fig. 2. Methodology of Recommender System
tomer in selecting the best product during the search process.
The Naive Bayesian FrameWork was utilized in [16] to effec-
A. Modelling Preferences of User used as a part of predictive modeling in machine learning and
We look at user selections to gather data based on user data mining [23].
interests. This is where the activity log of a specific user has IV. R ESULTS
analyzed through the type of content that a user is interested Considering the mechanism, it can produce the results
in. This is a user-centered approach for data collection relevant by applying different recommendation functions to generate
to user interests. a desired recommendation. We separated our dataset into
1) Prediction: To create an apparel recommendation list, training data and test data. 80% of the data formed the
data that is relevant to user interests are sorted and further training data and the rest 20% was made the test data. The
used for making predictions. These predictions are based on training dataset contains 50 users, 100 ratings and 25 items.
the probability calculated with the help of similarity functions. Hence results are exclusively segregated into two major
courses: First of all, the quality of results for every single
B. Nearest Neighbor Algorithm recommendation function as well as the best performance
Among non-parametric classifiers, one of the most popular recommendation function. While color similarity extract from
non-parametric techniques in the group of the Nearest Neigh- content base filtering and recommendation over user purchase
bor (NN) [18]. Based on the supplied image, the closest neigh- history also filtered on the basis of content, as shown in Fig
bor algorithm is utilized to locate the most relevant products, 3.
and suggestions are provided [19]. Classifier is usually utilized
in exercise for the ones including the foundation of the sample-
to-sample span. It mainly utilized the distance function to find
out the parallelism within two different samples to intent the
nearest neighbor for testing a data or point of any query [20].
1) Content-based Apparel Recommendation System:
Content-Based Filtering recommender systems allow users to
filter information based on a collection of procedures that per-
form analysis to predict what the user could be interested in or
what is relevant to the user [21]. Content-based recommender
systems are designed to suggest goods that, are comparable to
those that have previously piqued a user’s interest. Content-
based recommendation systems can be used to propose web
pages, news articles, restaurants, television shows and objects
for sale, among other things [22]. In this research, Grab Now
Fig. 3. Results of the Recommender System
E-Commerce platform is considered as a run time environment
for application and evaluation of an apparel recommendation The overall evaluation can be summarized considering the
system. In order to design an apparel recommendation system, sensitivity and significance of few parameters. Content-based
we use profiles of apparel information (such as type and colors filtering approaches the algorithms utilizing nearest neighbor
of apparels) viewed by different users. Every single interest algorithm. It is accomplished by using the results from users’
shows a group of similar apparels with regard to its type experiments, gathered from database. Three major steps can
and colors. The prediction module is used for providing a be used for required evaluation. Every single algorithm out of
recommendation list to the users of Grab Now E-Commerce evaluation table in database gets created which contained the
platform. categories, item ID’s, item colors and importantly item name.
This mechanism executed algorithms based on two different Considering the previous step every item from created list,
similarities. The system relied on item-to-item Nearest Neigh- by using the proposed parameters producing evaluation, from
borhood algorithm that, is for user purchase history followed which neighbor item list can be created. Above process can
by recommendation of similar products and similar color in replicate by every single selection each time made by user
item. Either in the database or the item selected by the user, interference. User selection basically opt the recommendation
these two different degrees of similarities are computed under while the prediction is aligned on identical item attributes
the function of recommendation. Same database is used for which users selected by its own.
showing more accuracy in either of the one which is already We also experience the two different similarities, item color
tested for selecting items considering selecting neighbor. and recommendation of product through purchase history
of any user, where item to item similarity measures in
C. Naı̈ve-Bayesian Frame Work Nearest Neighbor algorithm one has far better advantage over
Naı̈ve-Bayes is one of the most dominant frameworks another, that is color similarity is much more on a higher
used for prediction based on probability [19]. It collects side of recommendation instead of user purchase history then
observations regarding an item and calculates its value. This importantly the edge goes with the color similarity note.
classification tool is derived from Bayes’ Theorem. It is mostly
 V. C ONCLUSION [15] G. N. Kottage, D. K. Jayathilake, K. C. Chankuma, G. U. Ganegoda, and
T. Sandanayake, “Preference based recommendation system for apparel
With the expansion of e-commerce platforms and Internet e-commerce sites,” in 2018 IEEE/ACIS 17th international conference on
transactions, intelligent e-commerce personalized recommen- computer and information science (ICIS). IEEE, 2018, pp. 122–127.
[16] S. Bandyopadhyay, S. Thakur, and J. Mandal, “Product recommendation
dation systems have made significant progress in recent years. for e-commerce business by applying principal component analysis (pca)
Faced with such vast amounts of user data, establishing an and k-means clustering: benefit for the society,” Innovations in Systems
intelligent recommendation system based on user interests is and Software Engineering, vol. 17, no. 1, pp. 45–52, 2021.
[17] J. S. Mohan, H. S. Vedantham, V. C. Vanam, and N. P. Challa, “Product
critical for making human lives easier. With an abundance recommendation systems based on customer reviews using machine
of already existing information and data on the web it is learning techniques,” in Data Intelligence and Cognitive Informatics.
extremely difficult for the users to find relevant data. This Springer, 2021, pp. 267–286.
[18] Y. Zeng, Y. Gong, and X. Zeng, “Controllable digital restoration
issue is solved by recommending new and not-yet applied of ancient paintings using convolutional neural network and nearest
options to users based on their current requirements and wants. neighbor,” Pattern Recognition Letters, vol. 133, pp. 158–164, 2020.
To provide clear suggestions, recommendation systems use [19] M. Sridevi, N. ManikyaArun, M. Sheshikala, and E. Sudarshan, “Person-
alized fashion recommender system with image based neural networks,”
various types of information based on user preferences, lists of in IOP Conference Series: Materials Science and Engineering, vol. 981,
items accessible and the history of previous searches between no. 2. IOP Publishing, 2020, p. 022073.
the user and the systems used. [20] E. P. Frady, G. Orchard, D. Florey, N. Imam, R. Liu, J. Mishra, J. Tse,
A. Wild, F. T. Sommer, and M. Davies, “Neuromorphic nearest neighbor
R EFERENCES search using intel’s pohoiki springs,” in Proceedings of the Neuro-
inspired Computational Elements Workshop, 2020, pp. 1–10.
[1] S. E. Küçükbay and N. K. Turhan, “Supervised learning based apparel [21] B. Hrnjica, D. Music, and S. Softic, “Model-based recommender sys-
recommendation system for online shopping,” ISC2019, 2019. tems,” Trends in Cloud-based IoT, pp. 125–146, 2020.
[2] Q. Zhang, J. Lu, and Y. Jin, “Artificial intelligence in recommender [22] M. J. Pazzani, D. Billsus et al., “The adaptive web,” Berlin, Heidelberg:
systems,” Complex & Intelligent Systems, vol. 7, no. 1, pp. 439–457, Springer-Verlag, pp. 325–341, 2007.
2021. [23] H. Zhang, L. Jiang, and L. Yu, “Attribute and instance weighted naive
[3] M. Naumov, D. Mudigere, H.-J. M. Shi, J. Huang, N. Sundaraman, bayes,” Pattern Recognition, vol. 111, p. 107674, 2021.
J. Park, X. Wang, U. Gupta, C.-J. Wu, A. G. Azzolini et al., “Deep
learning recommendation model for personalization and recommenda-
tion systems,” arXiv preprint arXiv:1906.00091, 2019.
[4] T. Ebesu, B. Shen, and Y. Fang, “Collaborative memory network
for recommendation systems,” in The 41st international ACM SIGIR
conference on research & development in information retrieval, 2018,
pp. 515–524.
[5] P. Meel, P. Chawla, S. Jain, and U. Rai, “Hybrid apparel recommendation
system based on weighted similarity of brand and colour,” in Inter-
national Conference on Innovative Computing and Communications.
Springer, 2021, pp. 427–437.
[6] M. Vivek, N. Manju, and M. Vijay, “Machine learning based food recipe
recommendation system,” in Proceedings of international conference on
cognition and recognition. Springer, 2018, pp. 11–19.
[7] N. Nikzad-Khasmakhi, M. Balafar, and M. R. Feizi-Derakhshi, “The
state-of-the-art in expert recommendation systems,” Engineering Appli-
cations of Artificial Intelligence, vol. 82, pp. 126–147, 2019.
[8] Y. Lu, Y. Duan, and R. Saga, “Double deep features for apparel rec-
ommendation system,” in Proceedings of the 53rd Hawaii International
Conference on System Sciences, 2020.
[9] C. Guan, S. Qin, and Y. Long, “Apparel-based deep learning system
design for apparel style recommendation,” International Journal of
Clothing Science and Technology, 2019.
[10] X. Li and D. Li, “An improved collaborative filtering recommendation
algorithm and recommendation strategy,” Mobile Information Systems,
vol. 2019, 2019.
[11] S.-F. Tsarouchis, A. S. Vartholomaios, I.-P. Bountouridis, A. Karafyl-
lis, A. C. Chrysopoulos, and P. A. Mitkas, “Science4fashion: An
autonomous recommendation system for fashion designers,” in IFIP
International Conference on Artificial Intelligence Applications and
Innovations. Springer, 2021, pp. 729–742.
[12] L. Jiang, Y. Cheng, L. Yang, J. Li, H. Yan, and X. Wang, “A trust-
based collaborative filtering algorithm for e-commerce recommendation
system,” Journal of Ambient Intelligence and Humanized Computing,
vol. 10, no. 8, pp. 3023–3034, 2019.
[13] R. R. Gudur, A. Blackler, V. Popovic, and D. Mahar, “Methodological
issues with older users as research participants,” in Proceedings of
the 6th IASDR (The International Association of Societies of Design
Research Congress. IASDR (The International Association of Societies
of Design Research), 2015, pp. 851–867.
[14] H. Zarzour, Z. Al-Sharif, M. Al-Ayyoub, and Y. Jararweh, “A new
collaborative filtering recommendation algorithm based on dimension-
ality reduction and clustering techniques,” in 2018 9th international
conference on information and communication systems (ICICS). IEEE,
2018, pp. 102–106.

View publication stats