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

net/publication/356682233

A Review for Artificial Intelligence Solutions at E-commerce field

Conference Paper · June 2021

CITATION READS
1 514

2 authors, including:

Bahadır Akın Akgül

Galatasaray University
2 PUBLICATIONS 1 CITATION

SEE PROFILE

All content following this page was uploaded by Bahadır Akın Akgül on 01 December 2021.

The user has requested enhancement of the downloaded file.

 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

E-ticaret Alanındaki Yapay Zekâ Çözümleri Üzerine Bir İnceleme

Bahadır Akın AKGÜL1,2*, Prof.Dr. Kemal FİDANBOYLU1

1
Bursa Uludağ Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği, Bursa, TÜRKİYE
2
Hepsiburada.com Ar-Ge Merkezi, İstanbul, TÜRKİYE

Özet
E-ticaret, internet kullanımının yaygınlaşmasıyla birlikte, alışılagelmiş ticaret anlayışının
elektronik ortamlarda yapılabilmesini sağlamış, kullanıcıların hayatını kolaylaştıran
çözümlerle birlikte her geçen gün büyüyen, heyecan verici bir araştırma alanı haline
gelmiştir. Geleneksel ticaret anlayışının aksine ürünlerin sergilendiği ortamların elektronik
olması nedeniyle, e-ticaret alanındaki şirketlerin, akıllı çözümlerle pazar paylarını
arttırabilme çabasına yol açmıştır. Yakın dönemde, e-ticaret çözümlerinde kullanıcıların
hayatını kolaylaştıran yapay zekâ çözümlerine yaygın olarak kullanılmaya başlanmıştır.
Örneğin, sahte ürün yorumlarının tespit edilmesi, ürünlerin fiyatlarındaki değişimlerin ürün
talebine göre olarak yapılabilmesi, mükerrer olarak girişi yapılan ürünlerin tespit edilmesi,
ürünlerin kategorileme işlemlerinin gerçeklenmesi, ürün ile ilintili yeni ürün önerisi
yapılabilmesi ve ürün özelliklerinde yer alan tutarsızlıkların giderilmesi gibi uygulamalar
yapay zekâ çözümleri içermektedir. Bu bildiride, makine öğrenimi, doğal dil işleme, yapay
sinir ağları gibi yapay zekâ yaklaşımlarının e-ticaret alanındaki uygulamaları incelenmiştir.
Anahtar Kelimeler: E-ticaret, Yapay Zekâ, Makine Öğrenmesi, Doğal Dil İşleme, Yapay Sinir Ağları

A Review for Artificial Intelligence Solutions at E-commerce field

Abstract
With the widespread use of the internet, e-commerce has enabled the traditional
understanding of commerce to be carried out in electronic environments, and has become
an exciting and growing research area with solutions that make users' lives easier. Contrary
to the traditional understanding of trade, since the environments where the products are
exhibited are electronic, it has led companies in the field of e-commerce to increase their
market shares with smart solutions. Recently, artificial intelligence (AI) solutions that
facilitate users' lives in e-commerce solutions have been widely used. For example,
applications such as detecting fake product comments, making changes in the prices of
products according to product demand, identifying repeatedly entered products, realizing
product categorization, making new product suggestions related to the product, and
eliminating inconsistencies in product features include AI solutions. In this paper, the
applications of AI approaches such as machine learning, natural language processing and
artificial neural networks (ANN) in the field of e-commerce are examined.
Keywords: E-commerce, Artificial Intelligence, Machine Learning, NLP, ANN

1 Introduction platforms that provide the sale of daily basic

Electronic commerce, e-commerce or online products are also becoming widespread day by day,
shopping is an area that has grown enormously especially due to the restrictions brought by the
through platforms that mediate the sale of many Covid-19 pandemic.
products on the internet. Today, besides electronic Usually, an ideal e-commerce site consists of pages
products such as mobile phones and televisions, listing different products. These pages usually
bags, shoes and clothes are sold through include product features, product photos,
e-commerce sites. In addition, e-commerce information about the seller of the product,

1
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

information on how to deliver the product, and the entire graphics created by professional
comments made by users who have used this designers. Finally, they applied Thompson
product before. Sampling through variational inference for efficient
Different sellers may sell the same products on exploration to decrease the large variance caused
e-commerce sites in the marketplace model such as by restricted feedback, and they executed an
Amazon.com or Hepsiburada.com. Users who want automated interaction function search to improve
to purchase products with the help of the features the expressiveness of the model [1].
on their websites can request information about To improve the impacts of product promotion and
the products directly from the product vendor. the likelihood of advertisement clicks, internet
In this paper, artificial intelligence (AI) applications advertising organizations desire to increase
in e-commerce are explained under seven visitors' website browsing duration/time on site.
headings. Namely: AI solutions for product For measuring the time spent on a website, a group
advertising, marketing, and sales; AI solutions for of researchers presented a hybrid prediction model
finance; customer-related AI solutions; AI solutions that combines self-organizing map (SOM) and long
for demand and price; e-commerce product and AI short-term memory (LSTM) [2].
solutions; warehouse and logistics. 2.2 Marketing
2 AI Solutions for Product Advertising, Using AI solutions to market a product is a good
Marketing, and Sales way to stand out from competitors in e-commerce.
Although e-commerce websites sell products in Precision marketing entails segmenting the market
different areas such as technology, food, clothing and consumers using the findings of big data
and health, their main purpose is to generate more analysis, the Internet, and big data mining
income. Many people work in the field of sales and technologies, and having focused and in-depth
marketing in e-commerce companies that sell their engagement with segmented consumers to
own products. In addition, e-commerce sites in the understand their demands, in order to push them
marketplace model, where other sellers' products to the most required products and services at the
are also sold, also help these sellers sell and market most appropriate moment. The identification of
their products with their own teams. Advertising is target audiences, the selection of communication
one of the best ways to increase sales. E-commerce techniques, and the measuring of marketing results
websites enable product ads to be displayed with are the three primary processes in precision
Google ads as well as on-site advertising solutions. marketing [3].
To thoroughly anticipate sales, a group of
2.1 Advertising
researcher employed retail household data and the
Advertising has an important role in selling k-means algorithm to subdivide the value of retail
products on e-commerce sites. The purpose of households; the Boston matrix is used to categorize
these ads is to influence the user with attractive the market; and the ARMA model, Holt-Winter
media and make the user click on the ad. model, and Delphi expert survey technique are
Click-through rate (CTR) is a metric that measures combined [3]. In addition, another study proposed
how many clicks advertisers receive on their ads a solution using K-means++ with random forest
compared to how many impressions they receive. (RF) for precision marketing. They offer a
CTR increases can be seen as a signal of increased clustering approach that combines related criteria
income for merchants and e-commerce businesses with multi-valued discrete features. (MDF) [4].
[1]. Motivations such as budget constraints and
reaching more users have paved the way for the 2.3 Sales
use of AI solutions in e-commerce product Eventually, the ultimate goal targeted with
advertisements. advertising and marketing is sales.
For instance, a solution that offers one-shot search In another study, convolutional neural networks
strategies for quickly finding efficient interaction (CNN) were used for mining e-commerce data in
functions between creative elements was order to anticipate commodities sales. They have
developed to predict CTR with help of Automated used the intrinsic attribute information of
Machine Learning (AutoMT). Only composited commodities to transform the original log data
creatives were examined in this study, rather than connected to commodities into a specific "data

2
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

frame" format. Experiments on huge data sets instance, in a study that used RF algorithms to
suggested that the proposed approach may detect customer transaction frauds, they proved
enhance the accuracy of sales forecasting that RF is better than SVM and LR algorithms [13].
significantly [5]. Furthermore, in another research study, credit card
frauds were detected by using the CRED algorithm
3 AI Solutions for Finance
which is a continuous/discrete rule extractor via
In order to purchase an item through e-commerce DT induction [14].
websites, the required fee must be paid with a
payment method. The payment can be received 4 Customer-Related AI Solutions
with many different methods such as credit card, People who buy products listed on e-commerce
payment on product delivery, open bill payment or websites are called customers. In e-commerce sites
direct bank account payment. In particular, in the Business to Customer (B2C) model, the
methods other than credit card payment pose a customer is the traditional user. On the other hand,
credit risk in terms of collection of the payment. in e-commerce sites in the Business to Business
(B2B) model, the customer is a business.
3.1 Credit Risk
E-commerce websites divide their customers into
Credit Risk Forecasting (CRF) is a tool that helps different groups according to their various
determine whether or not a buyer should be characteristics. In this way, they can apply different
accepted, as well as the discount coefficient when approaches to different groups and this process is
determining credit limits. called customer segmentation. However, in
Some researchers analyzed the relationship addition to customer segmentation, more detailed
between CRF and credit limit setting tasks in a information is required. This required information
financial scenario and suggested MvMoE, a unique is provided by customer behavior identification
multi-task learning neural network that can methods, and this information helps the
improve both tasks at the same time [6]. For credit e-commerce website to know its customers better.
risk evaluation, artificial neural network (ANN), On the other hand, customers can share their
decision tree (DT) and logistic regression (LR), and experiences with the e-commerce sites.
DT-dynamic bayesian network (DBN) approaches 4.1 Segmentation
were used [7]. Moreover, the particle swarm
algorithm (PSO), support vector machine (SVM) Customer segmentation can be performed over
based e-commerce mobile payment risk prediction different parameters such as the amount of
model was trained based on a large number of expenditure made on the site, the rate of
gathered risk prediction sample reference data in cancellation/return/change in the orders made.
order to minimize the occurrence of probable risks For instance, a research study aimed to analyze
[8]. A deep learning-based financial risk early two-year customer data from 2018 to 2019 of a
warning model was also proposed for a successful company and suggested a k-means algorithm for a
use of deep learning in corporate governance [9]. better customer segmentation [15]. In another
research study, researchers have worked on
3.2 Fraud
segmenting an e-commerce website data by
Using SVMs and Risk Forecasting (RF) classifiers, applying the regularized k-means method instead
some researchers presented a credit card fraud of standard k-means and they proposed a novel
detection system based on operational and elastic new penalty method for k-means [16]. In
transaction data [10]. Some other researchers addition to the above, other researchers have
presented a solution based on the XGBoost proposed a neural network-based data mining
algorithm for detecting fraud users [11]. In method (FuzzyART) to help with market targeting
addition, a quantum behavior-based particle [17]. Also, a combination of FP-Grow algorithm and
swarm optimization approach with global search Naive Bayesian (NB) algorithm which applied user
capabilities, which is integrated with the fast segmentation was proposed by some researchers
converging k-means approach to boost efficiency [18]. Furthermore, some researchers concentrated
and efficacy, was proposed in a study for on the detection of customer quality with
discovering collaborative malevolent network perception (MLP) and ANN modeling [19].
behavior in e-commerce [12]. Also, customer
transaction data was used to detect fraud. For

3
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

4.2 Behavior Prediction four different algorithms and suggested that the
As mentioned earlier, predicting customer KNN gives the best result [29].
behavior provides more information to be used As stated previously, some websites do not require
with customer segmentation data. the purchase of the product for commenting on the
A study aimed at explaining online shopping product. Although this increases the number of
intentions in Colombia with the help of NB comments made on the product, it also enables
algorithm and they used parameters like false comments to be made.
acceptance of internet, internet use frequency, and A study measured three different algorithms for
purchase intention [20]. Also, work has been done detecting fake reviews from a hotel review dataset
to detect churn customers. Usually, e-commerce and proposes NB algorithm as the most successful
websites define a customer churn when a user one [30]. Another work proposed an Ensemble
doesn't shop for a while. A work used data mining Model with stacking classifiers and claims that
methods including k-nearest neighbors (KNN) works better than KNN, RF and NB algorithms for
algorithm, NB, DTs, RF, Gradient Enhancement detecting spam reviews [31]. Also, a study claimed
Trees, and Induction Rules to detect churn that a combination of CNN and LSTM algorithms
customers [21]. Additionally, a work used a give more accurate results to detect fake reviews
combination of linear model LR and DT based [32]. In another study, NB and RF algorithms were
XGBoost model to predict customer buying used for detecting fake reviews and deleting them
behavior [22]. On the other hand, a study used a automatically [33].
hybrid model of Gradient Boosting DT and LR for
5 AI Solutions for Demand and Price
predicting customer buying behavior [23].
Unlike traditional commerce, e-commerce websites
4.3 Product Reviews are thought to offer an unlimited number of
Reviews written by customers for a product or products to list and purchase, but in reality this is
service sold on an e-commerce site are called not the case. Especially on days like eleveneleven
product reviews. Product reviews often include an or black friday, when many users flock to the site,
area where users can rate one to five, an area to the traffic of the website increases considerably
post their comments, and an area to add and problems may occur due to the density.
multimedia. Although most e-commerce sites However, it is important to be able to predict
require you to have purchased a product before customer traffic and demand not only on discount
you can comment on it, there are also websites that days but also on other days. In addition, it is very
show the opposite. In addition, some e-commerce useful to be able to predict the price dynamically in
sites may present your comments to users by order to prevent price anomalies and to prevent
passing an approval mechanism, even if you the constant updating of prices, especially in
purchase the product. e-commerce sites in the marketplace model.
For instance, a work targeting sentiment analysis 5.1 Traffic prediction & demand
of a set of reviews with three hybrid RF
It is important for e-commerce websites to be able
classification algorithms [24]. Also, a work has
to predict traffic, scale their systems and develop
been made with a continuous NB algorithm for
better performance systems. However, since
sentiment analysis [25]. Another study applied a
internet traffic is often irregular and unpredictable,
hybrid model that contains NB and LR [26]. Since
making accurate predictions is difficult. On the
the product comments in the studies can be in
other hand, some studies have been carried out on
different languages, there may be differences in the
this subject in countries such as China where all
application in specific languages. Therefore, some
internet traffic can be controlled.
studies have been done on languages other than
English. For instance, a study proposed to use the For instance, a work proposed a prediction
Gated Recurrent Unit (GRU) algorithm that is very algorithm with an updated Moth-Flame
similar to LSTM algorithm to analyse sentiments of Optimization (MFO) model with help of China
product reviews on Turkish language [27]. Besides, Internet Network Information Center dataset [34].
another study applied CNN and Recurrent Neural Similar to traffic prediction, a study proposed a
Networks (RNN) models for sentiment analysis transaction anomaly detection using an
[28]. Another study in Turkish language compared

4
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

e-commerce transaction dataset by K-means structure. Selling a product in the wrong category
algorithm [35]. makes it difficult for the user to reach the product
as well as reducing search and suggesting
5.2 Price
performance. For this reason, solutions have been
Price is the most important competitive factor developed that determine the category that the
especially for e-commerce sites in the marketplace product should belong to, using certain features of
model. It is the seller's responsibility to determine the product. For instance, a study proposed a
the price and has to give the best price to compete solution for categorization of products with help of
with other competitors. On the other hand, it is product title information and using gcForest
very useful to be able to make price estimation by a algorithm [40]. An approach also used product
system in offer-based products or services where titles in Myanmar language and applied different
there is no fixed price. algorithms like SVM, RF, NB and LR to detect the
A study proposed fuzzy association rule mining category of product [41]. Another approach using
and the fuzzy logic technique called Smart-Quo to B2B e-commerce website product title data and
predict a price for offer based services [36]. On the using NB algorithm was also proposed [42]. On the
other hand, a study used the Stack Generalization other hand, an approach used product description
method with linear regression, RF regression, to detect product categories with a solution
gradient boosting and DT regression algorithms to containing word2vec, paragraph2vec, CNN
predict price on a Turkish e-commerce website algorithms [43]. Another approach used title,
data [37]. Also, a method has been proposed to description and image information to detect
predict the price of mobile phones with a product categories by using Ensemble Model [44].
combination of LSTM neural network and Support Also, a study that used product information and
Vector Regression (SVR) algorithms [38]. Also, a applied GRU sequence-to-sequence model to real
hybrid algorithm that contains Least Squares SVM world e-commerce data [45]. In addition, work has
and Artificial Bee Colony (ABC) algorithms been done with a combined seq2seq method and
proposed to predict prices in a study [39]. uses product title and description to detect product
category [46]. Also, a study proposed a specified
6 E-commerce product and AI solutions
RNN algorithm to detect categories using product
In an e-commerce site, the structures that contain metadata [47]. In another approach, Multi-LSTM
many fields such as price, description, comments, and Multi-CNN were used for categorization and
photos, attributes and present them to the user in a attributes such as title, short description,
single model are called products. Especially in description, synopsis and brand were also utilized
e-commerce sites in the marketplace model, since [48]. Some other researchers have used title,
the same products can be created by different company name description and model number
users, there are product catalogs containing these features to apply categorization with help of NB
products. Whether it is in the product catalog or algorithm [49].
not, the attribute values must be of high quality in
order for a product to attract the attention of the 6.2 Enrichment
user and provide the right information to the user. When entering a product on an e-commerce site,
The process of improving product properties is many values such as product title, product
called enrichment, and stating that these values are description, product image are required to be
correct is called verification. In addition, there are entered. On the other hand, it is quite common for
very high quantities of products, especially on products in some categories to require certain
global and large e-commerce websites. In order for attribute values, such as brand. As the number of
users to find these products easily, there is a search features for product entry increases, the number of
feature on the website and the products are listed products that can be added to the system
under some predetermined categories. In addition, decreases. For this reason, various systems have
products that the user may want to buy are also been developed in order to improve the product
recommended on the website. quality and to diagnose some features of the
product.
6.1 Categorization
For instance, a study used product title and
In e-commerce sites, it is mostly kept under
product description to extract attributes with
categories that are designed as a product tree
word2vec, Conditional Random Fields (CRF) and

5
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

RNN on an e-commerce website data [50]. There is SVM, RF, Model-Agnostic Meta-Learning (MAML),
also another work that aims to extract attributes and Meta-SGD algorithms [60]. Products that were
from product title and product description by using previously in the product catalogs can be added to
word2vec, TF-IDF, Distributed RF and Gradient the system again. In order to prevent this situation
Boost Machine (GBM) [51]. One of the ways to that may create duplicate products in the product
enrich the product is to show the highest quality catalog, systems have been designed to display
and attractive product images in the first place. A matching products to the user. A study proposed
study offered an image enrichment that includes Bidirectional Encoder Representations from
features like image ordering and selecting best Transformers (BERT) to achieve product matching
quality images with combinations of Inception-V3, with the help of a real world e-commerce website
Resnet-50, Deep embeddings, LR and RF catalog [61]. In another work, hybrid
algorithms [52]. bidirectional-LSTM and CNN algorithms were used
in achieving product matching which utilized a
6.3 Recommendation
dataset that contains product description, category
One of the best ways to improve the user type, title, brand and other attributes data of
experience is to improve the product products [62].
recommendation system. Making product
recommendations of high quality encourages the 6.5 Search
user to shop more, while allowing the e-commerce There are search features for users to find many
website to earn more. products on e-commerce sites easily. Just like a
For example, two studies offered the use of product search engine, search modules in e-commerce sites
images for product recommendation with the help also use smart systems today.
of deep CNN [53,54]. Additionally, a study used A study proposed a solution for zero result search
order transactions of products and using k-means result pages by analysing search queries of two
algorithm to detect similar products [55]. In different companies and proposes a related
another study, an extended version of Cat Swarm category with help of deep neural networks [63].
Optimization (CSO) with a product information Also, a work proposed a tail query classification by
dataset for recommending products was used [56]. using a few algorithms like KNN, CNN, LSTM for
Also, work was done by using product views, search query intent prediction [64]. Additionally, a
adding to cart and completing orders to implement study offered a search query clustering using
a product recommendation system using S-KNN Adaptive density-based spatial clustering
and VS-KNN [57]. Another work used brand (DBSCAN) and k-means algorithms [65].
information and order data of a specific time
7 Warehouse and Logistics
period to present a product recommendation
system by using k-means algorithm [58]. After creating a user order on an e-commerce site,
Furthermore, one of the research studies used a the process of delivering a physical product to the
recommendation algorithm skip-gram by utilizing user begins. This process consists of two parts:
negative sampling (SGNS) and k-means with help warehouse and shipping.
of a grocery shipping dataset and an e-commerce 7.1 Warehouse
order dataset [59].
Nowadays, the warehouses of e-commerce
6.4 Validation and Matching websites, where many products are sold, have
The feature information of a product is transferred become quite complex structures. AI solutions are
to the databases of e-commerce websites, often used to increase the efficiency in these complex
with the help of an interface or a file. However, structures. For instance, locations of products may
some errors may occur as this data is entered by a be redesigned to reduce time at order picking by
human. For example, if a word that the e-commerce help of k-means algorithm based clustering [66].
website prohibits from being included in the 7.2 Shipment
product description is entered in the product
The product ordered from the e-commerce website
description, the user should be warned.
is delivered to the customer through cargo
One of the research studies has used title and other companies. Cargo processes can also be improved
product attributes to achieve an automatic with AI solutions. For example, a study offers a
validation of a product using a combination of LR,

6
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

path optimization for e-commerce logistics by [9] Cao Y., Shao Y. & Zhang H. “Study on early warning
using a hybrid genetic algorithm [67]. of E-commerce enterprise financial risk based on
deep learning algorithm.”Electron Commer Res
8 Conclusions (2021).
In this paper, a compilation of different AI based [10] Sudha C., and Akila D., "Credit Card Fraud Detection
solutions offered in many different areas of System based on Operational & Transaction
features using SVM and Random Forest Classifiers,"
e-commerce such as sales, finance, customer,
2nd International Conference on Computation,
product, warehouse and shipping was presented. Automation and Knowledge Management (ICCAKM),
Even though they offer similar solutions, no 2021, pp. 133-138.
comparison was made about the related solutions, [11] Chin S.H., Lu C., Ho. P.T, Shiao Y.F., Wu T.J.,
it was only aimed to guide the researcher. Although “Commodity anti-counterfeiting decision in
the main focus was mainly on the current research e-commerce trade based on machine learning and
topics, some of the relevant past research was also Internet of Things”, Computer Standards &
presented. Interfaces, Volume 76, 2021, 103504, ISSN
0920-5489,
Acknowledgments [12] Zhou M., Han L., Lu H., Fu C., An D., “Cooperative
I would like to express my gratitude to malicious network behavior recognition algorithm
Hepsiburada.com for their support. in E-commerce”, Computers & Security, Volume 95,
2020, 101868, ISSN 0167-4048.
References [13] Shaohui D., Qiu G., Mai H. and Yu H., "Customer
[1] Chen J., Xu J., Gangwei J., Tiezheng, Ge Z.Z., Lian D., Transaction Fraud Detection Using Random Forest,"
and Zheng K., Automated Creative Optimization for IEEE International Conference on Consumer
E-Commerce Advertising. Proceedings of the Web Electronics and Computer Engineering (ICCECE),
Conference 2021 (WWW ’21), April 19–23, 2021, 2021, pp. 144-147.
Ljubljana, Slovenia. [14] Youssef B., F. and Brahim O., "Rules Extraction
[2] Kao, L.J., Chiu, C.C., Wang, H.J., Ko, C.Y. Prediction of Bouchra and Deep Learning for e-Commerce Fraud
remaining time on site for e-commerce users: A Detection," 6th IEEE Congress on Information
SOM and long short-term memory study. Journal of Science and Technology (CiSt), 2020, pp. 145-150.
Forecasting. 2021; 1– 17. [15] Punhani R., Arora V., Sai Sabitha A., & Shukla V.,
[3] Cui, F., Hu, H. & Xie, Y. An intelligent optimization “Segmenting e-Commerce Customer through Data
method of E-commerce product marketing. Neural Mining Techniques.” Journal of Physics: Conference
Comput & Applic 33, 4097–4110 (2021). Series, 1714, 012026,2021.
[4] Zhang B., Wang L., Li Y., "Precision Marketing [16] H. -H. Zhao, X. -C. Luo, R. Ma and X. Lu, "An Extended
Method of E-Commerce Platform Based on Regularized K-Means Clustering Approach for
Clustering Algorithm", Complexity, vol. 2021. High-Dimensional Customer Segmentation With
[5] Pan, H., Zhou, H. Study on convolutional neural Correlated Variables," in IEEE Access, vol. 9, pp.
network and its application in data mining and 48405-48412, 2021.
sales forecasting for E-commerce. Electron Commer [17] Park S., "Neural networks and customer grouping in
Res 20, 297–320 (2020). e-commerce: a framework using fuzzy ART,"
[6] Liang T., Zeng G., Zhong Q., Chi J., Feng J., Ao X., and Proceedings Academia/Industry Working Conference
Tang J., “Credit Risk and Limits Forecasting in on Research Challenges '00. Next Generation
E-Commerce Consumer Lending Service via Enterprises: Virtual Organizations and
Multi-view-aware Mixture-ofexperts Nets.” Mobile/Pervasive Technologies. AIWORC'00. (Cat.
Proceedings of the Fourteenth ACM International No.PR00628), 2000, pp. 331-336.
Conference on Web Search and Data Mining (WSDM [18] Rao H., Zeng Z. and Liu A., "Research on
’21), March 8–12, 2021, Virtual Event, Israel. personalized referral service and big data mining
[7] Xu Y.Z., Zhang J.L., Hua Y, Wang L.Y., “Dynamic Credit for e-commerce with machine learning," 4th
Risk Evaluation Method for E-Commerce Sellers International Conference on Computer and
Based on a Hybrid Artificial Intelligence Model.” Technology Applications (ICCTA), 2018, pp. 35-38.
Sustainability. 2019; 11(19):5521. [19] Ballestar M.T., Grau-Carles P., and Sainz J.
[8] Guan X., Xie S.F., Liu F., Zhao H., and Liang Z., “Risk “Predicting customer quality in e-commerce social
Prediction in E-commerce Mobile Payment Based networks: a machine learning approach.” Rev
on PSO-SVM.” Proceedings of the International Manag Sci 13, 589–603 (2019).
Conference on Bioinformatics and Intelligent [20] Dakduk S., Horst E., Santalla Z., Molina, G., and
Computing (BIC 2021). Association for Computing Malavé J., “Customer Behavior in Electronic
Machinery, New York, NY, USA, 208–213. Commerce: A Bayesian Approach.” Journal of

7
 4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

theoretical and applied electronic commerce System," 6th International Conference on Inventive
research, 12(2), 1-20, 2017. Computation Technologies (ICICT), 2021, pp.
[21] Raeisi S. and Sajedi H., "E-Commerce Customer 985-988.
Churn Prediction By Gradient Boosted Trees," 10th [34] Zhang B., et al. “Forecasting of E-Commerce
International Conference on Computer and Transaction Volume Using a Hybrid of Extreme
Knowledge Engineering (ICCKE), 2020, pp. 055-059. Learning Machine and Improved Moth-Flame
[22] Liu C., Huang T.S., Ho P.T., Huang J.C., Hsieh C.T., Optimization Algorithm.” Applied Intelligence, vol.
“Machine learning-based e-commerce platform 51, no. 2, Feb. 2021, pp. 952–65.
repurchase customer prediction model.” 2020, PLOS [35] Tan X.S., Yang Z., Benlimane Y., and Liu E., "Using
ONE 15(12). Classification with K-means Clustering to
[23] Chen X., Ding S., Xiang Y., & Liu L. “Research on Investigate Transaction Anomaly," IEEE
prediction of online purchasing behavior based on International Conference on Industrial Engineering
hybrid model.” Journal Of Physics: Conference Series, and Engineering Management (IEEM), 2020.
2021, 1827(1), 012075. [36] Dai J., Wang T. and Wang S., "A Deep Forest Method
[24] Maganti S., , and Nalini N.,. “A Filter Based Improved for Classifying E-Commerce Products by Using Title
Decision Tree Sentiment Classification Model for Information," International Conference on
RealTime Amazon Product Review Data.” Computing, Networking and Communications
International Journal of Intelligent Engineering and (ICNC), 2020, pp. 1-5.
Systems, vol. 13, no. 1, Feb. 2020. [37] Tugay R., and Oguducu S.G., “Demand Prediction
[25] Alsubari, S.N., et al. “Development of Integrated Using Machine Learning Methods and Stacked
Neural Network Model for Identification of Fake Generalization.” ArXiv:2009.09756 [Cs, Stat], Sept.
Reviews in E-Commerce Using Multidomain 2020.
Datasets.” Applied Bionics and Biomechanics, edited [38] Chniti G., Bakir H. and Zaher H., E-commerce Time
by Fahd Abd Algalil, vol. 2021, Apr. 2021, pp. 1–11. Series Forecasting using LSTM Neural Network and
[26] Dadhich A., and Thankachan B., “Opinion Support Vector Regression. In Proceedings of the
Classification of Product Reviews Using Naïve International Conference on Big Data and Internet of
Bayes, Logistic Regression and Sentiwordnet: Thing (BDIOT2017).
Challenges and Survey.” IOP Conference Series: [39] Shahrel Z., Mutalib M., and Abdul-Rahman S.,
Materials Science and Engineering, vol. 1099, no. 1, “PriceCop – Price Monitor and Prediction Using
Mar. 2021, p. 012071. Linear Regression and LSVM-ABC Methods for
[27] Santur, Y., “Sentiment Analysis Based on Gated E-commerce Platform.” International Journal Of
Recurrent Unit.” International Artificial Intelligence Information Engineering And Electronic Business,
and Data Processing Symposium (IDAP), IEEE, 2019, 2021, 13(1), 1-14.
pp. 1–5. [40] Dai J., Wang T., and Wang S., "A Deep Forest Method
[28] Mabrouk, A. et al. “SEOpinion: Summarization and for Classifying E-Commerce Products by Using Title
Exploration of Opinion from E-Commerce Information," International Conference on
Websites.” Sensors, vol. 21, no. 2, Jan. 2021, p. 636. Computing, Networking and Communications
[29] Rumelli M., Akkuş D., Kart Ö. and Isik Z., "Sentiment (ICNC), 2020, pp. 1-5.
Analysis in Turkish Text with Machine Learning [41] Thant M., Yee K., and Nwet K.T., “Comparison of
Algorithms," Innovations in Intelligent Systems and Supervised Machine Learning Models for
Applications Conference (ASYU), 2019, pp. 1-5. Categorizing E-Commerce Product Titles in
[30] Rout, J.K., et al. “A Framework for Fake Review Myanmar Text.” International Conference on
Detection: Issues and Challenges.” International Advanced Information Technologies (ICAIT), IEEE,
Conference on Information Technology (ICIT), IEEE, 2020, pp. 194–99.
2018, pp. 7–10. [42] Lee, H.W., “지도학습 머신러닝 기반 카테고리 목록
[31] Suborna A.K., Saha S., Roy C., Sarkar S. and Siddique 분류 및 추천 시스템 구현.” 인터넷정보학회논문지,
M.T.H., "An Approach to Improve the Accuracy of vol. 20, no. 1, Feb. 2019, pp. 57–65.
Detecting Spam in Online Reviews," International [43] Ristoski, P., et al. “A Machine Learning Approach for
Conference on Information and Communication Product Matching and Categorization: Use Case:
Technology for Sustainable Development (ICICT4SD), Enriching Product Ads with Semantic Structured
2021, pp. 296-299. Data.” Semantic Web, edited by Claudia d’Amato, vol.
[32] Alsubari, S.N., et al. “Development of Integrated 9, no. 5, Aug. 2018, pp. 707–28.
Neural Network Model for Identification of Fake [44] Bi Y., et al. “A Multimodal Late Fusion Model for
Reviews in E-Commerce Using Multidomain E-Commerce Product Classification.”
Datasets.” Applied Bionics and Biomechanics, edited ArXiv:2008.06179 [Cs], Aug. 2020.
by Fahd Abd Algalil, vol. 2021, Apr. 2021, pp. 1–11. [45] Hasson I., et al. “Category Recognition in
[33] Anas S.M. and Kumari S., "Opinion Mining based E-Commerce Using Sequence-to-Sequence
Fake Product review Monitoring and Removal Hierarchical Classification.” Proceedings of the 14th

8
 View publication stats

4th International Conference on Data Science and Applications (ICONDATA’21), June 4-5, 2021, Istanbul, TURKEY

ACM International Conference on Web Search and at bol.com. In International Conference on Data
Data Mining, ACM, 2021, pp. 902–05. Engineering (ICDE), 2021.
[46] Tan L., et al. “E-Commerce Product Categorization [58] Bandyopadhyay S., et al. “Product Recommendation
via Machine Translation.” ACM Transactions on for E-Commerce Business by Applying Principal
Management Information Systems, vol. 11, no. 3, Component Analysis (PCA) and K-Means Clustering:
Aug. 2020, pp. 1–14. Benefit for the Society.” Innovations in Systems and
[47] Ha J.W., et al. “Large-Scale Item Categorization in Software Engineering, vol. 17, no. 1, Mar. 2021, pp.
e-Commerce Using Multiple Recurrent Neural 45–52.
Networks.” Proceedings of the 22nd ACM SIGKDD [59] Xu D., et al. “Theoretical Understandings of Product
International Conference on Knowledge Discovery Embedding for E-Commerce Machine Learning.”
and Data Mining, ACM, 2016, pp. 107–15. Proceedings of the 14th ACM International
[48] Krishnan A., and Amarthaluri A., “Large Scale Conference on Web Search and Data Mining, ACM,
Product Categorization Using Structured and 2021, pp. 256–64.
Unstructured Attributes.” ArXiv:1903.04254 [Cs, [60] Wang Y., et al. “Automatic Validation of Textual
Stat], Mar. 2019. Attribute Values in E-Commerce Catalog by
[49] Kim Y.G., et al. “Modified Naïve Bayes Classifier for Learning with Limited Labeled Data.” Proceedings of
E-Catalog Classification.” Data Engineering Issues in the 26th ACM SIGKDD International Conference on
E-Commerce and Services, edited by Juhnyoung Lee Knowledge Discovery & Data Mining, ACM, 2020, pp.
et al., vol. 4055, Springer Berlin Heidelberg, 2006, 2533–41.
pp. 246–57. [61] Tracz I. “BERT-based similarity learning for product
[50] Martin R., et al. “Accurate Product Attribute matching.” In Proceedings of Workshop on Natural
Extraction on the Field.” IEEE 35th International Language Processing in E-Commerce (pp. 66–75).
Conference on Data Engineering (ICDE), IEEE, 2019, Association for Computational Linguistics, 2020.
pp. 1862–73. [62] Li J., et al. “Deep Cross-Platform Product Matching
[51] Kemal T.U., et al. “Multi-Class Categorization of in e-Commerce.” Information Retrieval Journal, vol.
User-Generated Content in a Domain Specific 23, no. 2, Apr. 2020, pp. 136–58.
Medium: Inferring Product Specifications from [63] Tagliabue J., et al. “How to Grow a (Product) Tree:
E-Commerce Marketplaces.” Intelligent and Fuzzy Personalized Category Suggestions for ECommerce
Techniques in Big Data Analytics and Decision Type-Ahead.” ArXiv:2005.12781 [Cs, Stat], May 2020.
Making, edited by Cengiz Kahraman et al., vol. 1029, [64] Zhang J., et al. “Modeling Across-Context Attention
Springer International Publishing, 2020, pp. For Long-Tail Query Classification in E-Commerce.”
247–56. Proceedings of the 14th ACM International
[52] Chaudhuri, A., et al. “A Smart System for Selection of Conference on Web Search and Data Mining, ACM,
Optimal Product Images in E-Commerce.” IEEE 2021, pp. 58–66.
International Conference on Big Data (Big Data), [65] Darwin R.P., and Pasha M.F., "Search Query
IEEE, 2018, pp. 1728–36. Clustering Comparation On E-Commerce Using
[53] Rajest S.S., et al. “Extracting Related Images from K-Means And Adaptive DBSCAN," 3rd International
E-Commerce Utilizing Supervised Learning.” Conference on Mechanical, Electronics, Computer,
Innovations in Information and Communication and Industrial Technology (MECnIT), 2020, pp.
Technology Series, Feb. 2021, pp. 34–46. 207-211.
[54] Addagarla S.K., and Amalanathan A., “E-SimNet: A [66] Lorenc A., and Burinskiene A., “Improve the Orders
Visual Similar Product Recommender System for Picking in E-Commerce by Using WMS Data and
E-Commerce.” Indonesian Journal of Electrical BigData Analysis.” FME Transactions, vol. 49, no. 1,
Engineering and Computer Science, vol. 22, no. 1, 2021, pp. 233–43.
Apr. 2021, p. 563 [67] Yan, D., and Wu P., “E-Commerce Logistics Path
[55] Pan H., and Yang X., “Fast Clustering Algorithm of Optimization Based on a Hybrid Genetic Algorithm.”
Commodity Association Big Data Sparse Network.” Complexity, edited by Wei Wang, vol. 2021, Mar.
International Journal of System Assurance 2021, pp. 1–10.
Engineering and Management, Mar. 2021.
[56] Sharma N., Shambhu, and Prasanna Sadagopan.
“Influence of Conditional Holoentropy-Based
Feature Selection on Automatic Recommendation
System in E-Commerce Sector.” Journal of King Saud
University - Computer and Information Sciences, Jan.
2021, p. S1319157820306303.
[57] Kersbergen B., & Schelter S., Learnings from a Retail
Recommendation System on Billions of Interactions