An E-Commerce Recommendation Systems Based on Analysis

of Consumer Behavior Models

Valentyna Pleskach 1,2, Oleksandra Bulgakova 1, Viacheslav Zosimov 1, Elena Vashchilina 1

and Inga Tumasoniene 2

1
Taras Shevchenko National University of Kyiv, Bohdan Hawrylyshyn str. 24, Kyiv, 04116, Ukraine
2
Vilnius Gediminas Technical University Saulėtekio al. 11, Vilnius, LT-10223, Lithuania

Abstract
The article presents some models and strategies of consumer behavior in electronic commerce
systems. An analysis of e-commerce recommender systems was also carried out, including
context-aware recommender systems on the example of the analysis of regions of Ukraine and
weather conditions, etc. The core of the novelty of the research is primarily related to hybrid
behavioral models, personalization, contextual recommendations, especially integrating
contextual information, such as location, time of day, or user intent, evaluation metrics, etc.
The impact of machine learning and artificial intelligence on e-commerce recommender
systems and e-commerce systems in general is also explored. The article also presents the
microservice architecture of an electronic store with recommender system API. The
architecture includes several key components, such as client-side APIs for managing orders,
products and carts, services for processing customer requests and interacting with databases,
and data processing technology using artificial intelligence.

Keywords 1
E-commerce, consumer behavior model, e-commerce recommendation systems, adaptive user
profiling

1. Introduction
The widespread popularity of the Internet and the implementation of web technologies worldwide
have led to an increasing number of people abandoning traditional shopping methods and opting for
online payments and purchases. As globalization, informatization, and computerization continue to
advance, commercial transactions are increasingly taking place on specialized web pages such as online
stores, e-commerce platforms, online showcases, online auctions, online trading platforms, etc.
All of these are participants in e-commerce ecosystems. An e-commerce ecosystem can be
understood as an area of activity where buyers and producers play complementary roles. Components
of an e-commerce ecosystem typically include: customers, sellers, online marketplace, online payment
system, logistics, marketing, data analysis, etc. Let's explore the essence of consumer behavior models
and modern approaches in using recommender systems of e-commerce platforms.
Consumer behavior models play a crucial role in modern e-commerce systems. These models help
companies understand how consumers make decisions and interact with online platforms, enabling
them to adapt their products, services, and marketing strategies according to consumer needs. One of
the most widely used consumer behavior models is the buyer decision-making process model, which
consists of five stages: 1) problem recognition, 2) information search, 3) evaluation of alternatives, 4)
purchase decision, and 5) post- purchase evaluation and follow-up. E-commerce systems can use this

International Scientific Symposium «Intelligent Solutions» IntSol-2023, September 27–28, 2023, Kyiv-Uzhhorod, Ukraine
EMAIL: [email protected] (A. 1); [email protected] (A. 2); [email protected] (A. 3); [email protected] (A. 4);
[email protected] (A. 5)
ORCID: 0000-0003-0552-0972 (A. 1); 0000-0002-6587-8573 (A. 2); 0000-0003-0824-4168 (A. 3); 0000-0001-6867-6216 (A. 4); 0000-
0003-2408-5371 (A. 5)
©️ 2023 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)

210
model to guide consumers through each stage of the decision-making process, providing relevant
information and recommendations at each step.
Another important model is the consumer value hierarchy, which defines various levels of value that
consumers seek in a product or service, such as functional, emotional, and social benefits. E-commerce
systems can employ this model to understand the values consumers prioritize and accordingly develop
their products and services. Additionally, e-commerce systems can leverage behavioral economics
theories to influence consumer behavior, such as nudging methods to encourage consumers to make a
specific choice or increase their willingness to pay.
Overall, models of consumer behavior are essential for modern e-commerce systems to optimize
their offerings, user experiences, and sales conversions by understanding how consumers think, feel,
and behave when making purchasing.
The specificity of e-commerce means that traditional models of consumer behavior in a regular store
cannot be used when buying online. This is primarily due to the absence of physical attributes of the
store, such as the cleanliness of the goods and the trading floor, convenience of entering and store
location, and emotional and psychological factors that affect subjective consumer opinions.
Additionally, trust factors related to e-stores, products, and transactions are of utmost importance in e-
commerce. As a result, consumer behavior models have undergone significant changes. The emergence
and spread of the Internet cannot be viewed in isolation, and the models discussed below are a synthesis
of offline and online behavior.
M. Wenzl's research identified significant e-commerce trends that drive online consumer behavior.
Convenience is a top priority, and online shoppers seek easy access across all devices, looking for a
seamless transfer between devices throughout their buying journey. Access to omnichannel shopping,
with multiple technology options, engages consumers in an ‘always-on’ shopping experience, and most
consumers now shop while multitasking. Effortless payment and fast and reliable delivery are also
must-have factors when shopping online [1].
In [2] authors explore the strategies that online shop owners can use to stimulate customers to
explore new products, suggesting that a combination of surprise, novelty, and interactivity can lead to
more exploratory behavior. An integrated model of e-service quality, customer value, and switching
costs to enhance consumer engagement in e-commerce websites, finding that these factors positively
impact customer engagement proposed in [3]. In [4] investigated the decision-making processes
underlying consumers' choices between same-day and next-day delivery in online shopping, finding
that factors such as product type, urgency, and situational characteristics play a role in this decision.
The concept of digital affordances in the context of business-to-consumer e-commerce, highlighting
the importance of considering consumers' perceptions of technology's capabilities when designing
digital platforms proposed in [5]. Also, in [6] was research the design of recommendation systems on
e-commerce platforms with multiple objectives, proposing a multi-objective optimization model that
considers the trade-offs between relevance, diversity, and novelty.
Overall, e-commerce research helps understand how to enhance the online shopping experience for
consumers, create more effective digital marketing strategies, and develop relevant recommender
systems for e-commerce. This is important as online commerce continues to grow and gain more
popularity. Technological approaches to understanding consumer behavior in e-commerce, such as the
utilization of technology adoption models and impulse buying models have been discussed in [7-10].
Many researchers study issues related to consumer behavior in e-commerce and propose various
models and frameworks for analyzing and predicting consumer behavior. Examples include a model
for analyzing consumer behavior in e-commerce considering service quality [11], a model for
evaluating the role of customer behavior history, product classification, and pricing in the success of
recommendation systems in e-commerce [12], a model for constructing hierarchical polarization
characteristics in conjunction with consumer behavior in social e-commerce [13], a framework for
evaluating online markets considering trust and sustainability to predict customer behavior in e-
commerce [14], as well as a model that combines virtual reality technology and gamification to enhance
the consumer experience in tourism e-commerce [15]. These models cover different aspects of
consumer behavior in e-commerce. How-ever, to create a comprehensive understanding, it is necessary
to consider the dynamics of digital markets and trends in their development, as well as important tools
for supporting and advancing e-commerce, such as recommendation systems. Recommender system of
e-commerce recommender system of e-commerce systems are algorithms and techniques used in online

211
shopping platforms to suggest relevant and personalized product recommendations to individual users.
These systems aim to enhance the user experience, increase customer engagement, and ultimately
improve conversion rates and sales. The primary goal of e-commerce recommender systems is to
present users with a set of items that match their preferences, interests, and past behavior.
These recommendations can be based on various data sources, including user profiles, browsing
history, purchase history, demographic information, and explicit feedback (such as ratings and reviews).
The system analyzes this data to identify patterns, similarities, and correlations among users and items.
Personalized recommendation refers to the practice of suggesting products, services, or content to
individuals based on their unique preferences, interests, and past behaviors. It aims to provide a tailored
and relevant experience to each user, increasing engagement, satisfaction, and ultimately driving sales
in the case of e-commerce. Implicit data, such as click pat-terns, browsing history, and time spent on
different pages, can also be collected; data collection - the system collects data on user interactions,
such as product views, purchases, ratings, and reviews.
This data is used to create a user profile or model that captures the user's preferences and behavior;
real-time adaptation - recommendation systems often update their recommendations in real time to
adapt to the user's changing preferences and behaviors. This can be achieved through continuous
monitoring and analysis of user interactions. It's important to note that implementing personalized
recommendation systems requires handling user data and privacy concerns responsibly. Companies
should be transparent about data usage, provide options for users to control their privacy set-tings, and
adhere to relevant data protection regulations.
Overall, personalized recommendation systems in e-commerce leverage user data and algorithms to
provide tailored product suggestions, enhancing the user experience and driving engagement and sales.
So, the objectives of this article are to investigate and analyze consumer behavior models and
recommendation systems in e-commerce within the dynamic digital environment of an ever-evolving
digital society. In this article, we will discuss the impact of artificial intelligence technologies for data
processing and machine learning in developing consumer behavior models and recommendation
systems in modern e-commerce systems.
Aspects of novelty in this research cover areas such as consumer behavior patterns, data analysis
techniques, and algorithms with personalization and customization. These models are based on factors
such as browsing history, past purchases, social interactions, demographics, personalization, etc.
The article proposes hybrid methods of data analysis to obtain meaningful information from
consumer behavior data. This involved applying machine learning, data mining, and statistical
techniques to identify patterns, trends, and correlations that can be used to provide accurate
recommendations. These algorithms can combine different techniques such as collaborative filtering,
content-based filtering or hybrid approaches to create personalized and relevant product
recommendations for individual users. This may to help explore ways to tailor recommendations based
on individual preferences, needs and context, providing a more tailored and satisfying user experience.
This may include measuring user satisfaction, conversion rates, click-through rates, or other relevant
metrics to validate the impact of proposed models and algorithms.
Overall, the novelty of the study is the combination of consumer behavior analysis, data mining
techniques and recommendation algorithms to create e-commerce recommendation system.

2. Materials and Methods

A qualitative research approach was employed in the study, which involved the analysis and
synthesis of data and a systematic approach. The results were then used to draw conclusions about the
significance of identified factors for the successful functioning of e-commerce recommender systems
and how consumer behavior models can be utilized to create effective digital marketing strategies and
personalized customer strategies. Overall, the materials and methods used in this research aimed to
provide a comprehensive understanding of key factors contributing to the success of e-commerce
systems, including descriptions of consumer behavior models and e-commerce recommender systems.

2.1. Models of Customer Behavior

There are many models that take into account the motives for product selection, the level of
awareness and independence in choice, the level of consumer satisfaction with the product, and the

212
direction and possibility of influencing consumer choice through marketing and advertising stimuli.
There are also other well-known consumer behavior models.
The rational model corresponds to the concept of an “economic man”, rationalizing their actions on
the path to personal gain. The irrational consumption model is based on behaviorism with the principle
of “stimulus-response”: if the emotional attractiveness of a product is presented to the consumer for a
sufficient amount of time in a comprehensible form, they will indeed begin to experience positive
emotions from its acquisition.
The motivated consumer model requires understanding the consumer's motives, reasons, and the
content of their attitudes in order to find means of stimulating their choice. The conformist consumer
model is an evolution of the previous model. In a consumption-driven society, the dominant attitudes
of individuals become those related to their desire to conform to the norms of social groups they want
to belong to. It forms their social identity through a specific consumer basket containing goods and
services that meet the group's standards.
The consumerist consumption model revolves around the idea of consumption itself. For individuals
in this model, the mere opportunity to consume becomes a significant symbol of achievement and life
fulfillment. The requirements that this model imposes on consumer behavior are high - these are the
requirements of environmental friendliness, safety, ethical purity of both the products themselves and
manufacturing companies. In the era of digital transformation, models of the information consumer
have emerged, whose decisions about choosing a product or service are completely tied to the electronic
devices that they actively use when choosing in digital markets. And this model has received significant
distribution [16-17]. In 2018, Forbes noted the growing popularity of the “Research Online / Purchase
Offline” model: people are increasingly studying offers on the Internet in detail, com-paring different
brands, and only then make a purchase decision [18].
A significant limitation in this model is that sometimes it is quite difficult for small and medium-
sized stores to track whether the study of a product and brand on the Internet really led to a further
purchase in an offline store. However, this consumer decision model should always be kept in mind
when branded advertising is launched on the Internet. Thus, you can understand how a launched
advertising campaign affects offline sales.
In addition to the behavior models described above, there is an integral model of consumer trust
according to the statement of L. Chong and a conceptual model of Citrine. L. Chong's approach is to
identify four reliability factors for online sellers: perceived integrity, competence, security control and
privacy control. The main focus of the Citrine model is the behavior of the consumer, which depends
only on the level of the consumer's knowledge of the Internet and computers in general, which is a
primary for the Internet consumer, indispensable condition and is taken into account initially.
Must remember, with the spread of Covid-19 some of the behavior models have ceased to be
relevant, while others have succumbed to significant changes. The coronavirus has changed consumer
behavior online. According to the [19] re-search, 18% of users began to consume video content more
often, 12% of people began to listen to music and radio more often, 15% of respondents began to use
online shopping services more often, and 76% of respondents noted, which changed the habits
associated with hobbies and content consumption after the introduction of the forced self-isolation
regime. Therefore, in the digital world, behavior models can be influenced by factors that have an
external impact on the user of the e-trade and can significantly change his behavior.
So, let's summarize some strategies of consumer behavior. Buyers in e-commerce employ various
personal strategies to enhance their online shopping experiences and achieve their desired outcomes.
There are some common strategies employed by e-commerce buyers. This in particular,
1. Research and comparison, buyers often conduct thorough research on products or services they
intend to purchase. They compare prices, features, reviews, and ratings across multiple e-commerce
platforms to find the best option.
2. Buyers rely on customer reviews and ratings to gain in-sights into the quality, reliability, and
performance of products. Positive reviews and high ratings can instill confidence in a purchase, while
negative feedback may lead buyers to consider alternative options.
3. Buyers often use wishlists or add items to their carts to save products they are interested in.
This allows them to review and compare options before making a final decision, and it also enables
them to track price changes or availability updates.

213
 4. Buyers may set up price alerts or use price tracking tools. These notifications inform them when
the desired item reaches a specific price point, helping them make cost-effective purchases.
5. Buyers actively search for discount codes, coupons, or promotional offers to reduce the overall
cost of their purchases. They may follow brands or retailers on social media, subscribe to newsletters,
or use dedicated coupon websites to find exclusive deal.
6. Security is a crucial concern in e-commerce. Buyers prioritize platforms that offer secure
payment options, such as encrypted transactions or trusted third-party payment processors, to safeguard
their personal and financial information.
7. Buyers review the re-turn and refund policies of the e-commerce platform or individual seller.
They seek hassle-free return options, reasonable refund terms, and clear instructions in case the product
does not meet their expectations.
8. When buyers have questions, concerns, or require additional information, they may reach out
to customer support via live chat (chat-bot on base AI), email, or phone. Prompt and helpful customer
service can positively influence their decision to proceed with a purchase.
9. Buyers keep track of the shipping status of their orders by utilizing tracking numbers or delivery
notifications. This allows them to estimate delivery times, plan accordingly, and ensure the package
arrives in a timely manner.
10. After completing a purchase, buyers often share their feedback and experiences through product
reviews or ratings. This not only helps other buyers make informed decisions but also provides valuable
insights to sellers and platforms, etc.
These personal strategies may vary from buyer to buyer based on their preferences, ages, priorities,
and previous experiences in e-commerce activities.
Therefore, in the consumer behavior model presented by us, we have built upon the well-known
eCDP (Electronic Consumer Decision Process Model), which examines five stages of decision-making:
problem recognition, information search, alternative evaluation, purchase, and outcomes [20]. This
model is widely recognized and extensively used in consumer behavior research in e-commerce. It
provides a comprehensive framework that describes the fundamental stages of consumer decision-
making in online purchases. The eCDP model offers insights into the key factors influencing consumer
decision-making and can serve as a foundation for developing digital marketing strategies and
personalized approaches for customers. Optimizing this model allows us to gain a deeper understanding
of how consumers make decisions in online purchases and how recommendation systems can be
tailored for optimal consumer impact, Figure 1.
So, at the first stage of “problem recognition” in accordance with the well-known eCDP model, we
propose incorporating pseudo-identification of users (building a data-base of anonymous users and rules
based on their presence in web applications). The main advantages of this approach lie in the fact that
the identification model of anonymous users in software products can be utilized as a dynamic identifier
for automatically adapting the interface of an online store to the identified user [21].

2.2. Analysis of recommender systems of e-commerce

A recommendation system is a tool that uses a series of algorithms, data analysis and artificial
intelligence (AI) to make recommendations online. Recommender systems are used by e-commerce
sites to suggest products to their customers. As mentioned above recommender systems in e-commerce
are information filtering systems that provide personalized recommendations to users based on their
preferences, behaviors, and past interactions with the platform or similar users. These systems are
designed to help users discover relevant products or services that they might be interested in, thereby
improving their shopping experience and increasing engagement and sales for e-commerce businesses.
The importance of recommender systems in driving sales and enhancing user experience in e-
commerce. Recommender systems in e-commerce play a crucial role in boosting sales and improving
user experience. These systems offer personalized recommendations based on user behavior and
preferences, leading to increased sales and customer satisfaction. By suggesting relevant products, they
enhance engagement and encourage repeat visits. Moreover, recommender systems excel at cross-
selling and upselling strategies, increasing the average order value. In platforms with vast catalogs, they
assist users in discovering desired items, ultimately driving more sales. Overall, effective recommender

214
systems provide a competitive edge, attracting new customers and retaining existing ones, resulting in
enhanced user experience and increased revenue.
The different types of recommender systems used in e-commerce. In e-commerce, there are several
types of recommender systems commonly used to provide personalized recommendations to users.
These systems leverage various techniques and algorithms to analyze user behavior, preferences, and
item characteristics.
Let's note there are several types of e-commerce recommender systems.
Content-based filtering recommends items to users based on item characteristics and user
preferences.

Figure 1: Model of customer behavior with adaptation mechanism and recommender system block
User-based collaborative filtering. It finds users who have similar preferences to the target user and
recommends items that those similar users have liked or purchased. Item-based collaborative filtering:
it identifies similar items based on user preferences and recommends items that are similar to the ones
the user has already interacted with.
Hybrid recommender systems. These systems combine multiple recommendation techniques, such
as content-based filtering and collaborative filtering, to provide more accurate and diverse
recommendations. Hybrid approaches aim to leverage the strengths of different methods and overcome
their individual limitations. For such purposes, approaches based on machine learning (ML) are also
appropriate.
Reinforcement learning uses ML algorithms to learn user preferences and optimize
recommendations over time. It involves continually adapting and refining the recommendation strategy
based on user feedback and interactions.
Knowledge-based recommender systems use domain-specific knowledge and rules to generate
recommendations.
Demographic-based filtering recommends items based on demographic information such as age,
gender, location, or occupation.

215
 Context-aware recommender systems consider contextual factors like time, location, and user
situations for personalized recommendations. Data enrichment can include a variety of parameters that
provide additional information about the context of transactions and the impact of external factors on
consumers:
 Geographical data, meaning the location of transactions, including city, country, address,
latitude and longitude, allows you to understand how regional characteristics affect purchases.
 Temporal data, such as date and time of transactions, helps identify trends across different time
periods.
 Weather conditions such as temperature, humidity, wind speed, precipitation, visibility and
cloud cover reveal the impact of weather on consumers and their purchasing decisions.
 The phases of the moon indicate the possible impact of cyclical changes on consumption.
Environmental data, such as air pollution levels, water quality, and other environmental factors, can
influence consumer choices.
 Social data, such as demographics, information about local events or celebrations, consumer
trends and sentiments, helps in understanding the social influence on purchases.
Enriching data with these parameters provides more detailed and contextual in-formation about
transactions, which helps improve marketing strategies, personalize offers and improve customer
interactions in e-commerce. Environmental data enrichment is an important step in gaining a more
complete and contextual understanding of e-commerce transactions. To perform the enrichment, we use
the location and time from each transaction in the dataset.
Enriching the data environment also allows you to study correlations between different parameters
and transaction data. For example, you can analyze the correlation between geographic data, such as
the location of transactions, and the level of spending or the popularity of a certain product in a certain
region. This makes it possible to identify connections between the location of consumers and their
purchases.
Correlation between time data and purchases can also be explored. For example, you can study
whether there is a relationship between a date or a day of the week and the volume of sales of a certain
product or service. This can help identify seasonal trends, popularity on certain days or times of day,
and the impact of holidays or special events on consumption. Correlation can also be studied between
weather conditions and shopping. For example, research shows that weather factors such as temperature
or precipitation can influence the consumption of certain goods or services. This makes it possible to
forecast demand depending on weather conditions and take this into account when developing
marketing strategies.
Additionally, social data such as demographics or trends can be used to explore correlations between
these factors and purchases. For example, it is possible to study whether there is a relationship between
the age of consumers and their preferences for goods or services. This helps to make more informed
decisions about the target audience and to set up a more effective advertising strategy. Analyzing
correlations between different data parameters can also help reveal more complex relationships and
trends. For example, correlations between geographic data, social characteristics, and purchases can be
studied to gain a deeper understanding of the influence of cultural factors on consumption.
It is important to note that correlation does not always mean causation between variables. It indicates
a statistical relationship between them, but does not necessarily indicate the cause or direction of this
relationship. Data enrichment and analysis of correlations between various parameters allow increasing
the level of understanding and prediction of consumer behavior. This provides opportunities to develop
more effective marketing strategies personalize offers and improve customer interaction in e-
commerce. An analysis was conducted by region in Ukraine, and the following parameters were
determined: Average Purchase Value (APV), Average Purchase Frequency Rate (AP FR), Customer
Value (CV). The study showed interesting results indicating different levels of customer value in
different regions of Ukraine (Table 1).
Firstly, it was established that the highest lifetime customer value is observed in Odesa. This
suggests that customers from this region are more likely to make repeat purchases, remain interested in
the brand, and spend more on online shopping. On the other hand, the lowest customer lifetime value
was found in the Volyn region. This may indicate less activity and less frequent purchases by customers
from that region, or other factors affecting their value to the business.

216
 Analyzing average purchase value and purchase frequency will also allow companies to gain insight
into customer spending levels and customer activity across regions. Based on this data, pricing
strategies, bonus programs and other marketing initiatives can be developed to attract new customers
and retain existing ones.
Table 1
Analysis by regions of Ukraine.
Region of Ukraine APV AP FR CV
Odesa region 622.26 2.69 1670.88
Zaporizhia region 574.86 2.88 1652.94
Khmelnytsky region 704.58 2.32 1632.19
Kyiv region 608.45 2.64 1604.21
Donetsk region 641.18 2.45 1573.35
Kharkiv region 583.42 2.67 1558.51
Chernivtsi region 596.87 2.61 1558.14
Dnipropetrovsk region 606.99 2.55 1550.05
Kirovohrad region 611.29 2.51 1534.99
Luhansk region 640.54 2.35 1504.36
Mykolaiv region 600.74 2.46 1479.63
Ivano-Frankivsk region 600.60 2.44 1466.63
Ternopil region 630.96 2.32 1463.00
Rivne region 550.27 2.62 1441.66
Poltava region 598.11 2.41 1439.70
Zakarpattia region 634.79 2.25 1429.97
Zhytomyr region 595.70 2.31 1373.47
Kherson region 567.06 2.41 1367.96
Lviv region 581.70 2.34 1360.38
Vinnytsia region 636.21 2.13 1357.43
Cherkasy region 590.14 2.12 1249.69
Sumy region 561.29 2.09 1171.68
Chernihiv region 566.39 2.06 1168.19
Volyn region 594.89 1.94 1154.48
In addition, the analysis of the average purchase price and the frequency of purchases allows to
reveal the peculiarities of consumption in different regions. For example, if the average purchase price
is higher in one region, this may indicate the presence of customers with greater financial capabilities
or the presence of premium market segments. Such data can become the basis for the development of
special promotions or offers for target audiences.
The analysis showed that customer engagement increases by 14 percent during rain compared to
other weather conditions. This means that under the influence of rain, customers are more likely to
interact with businesses and make purchases.
Such analysis results indicate the importance of taking into account weather conditions and time of
day when developing marketing strategies. Businesses can use this information to target rain and
nighttime efforts by offering special offers, discounts or promotions that will attract more customers
and drive purchases. An analysis of attracting new regular customers based on external factors such as
time of day, weather conditions and atmospheric pressure was carried out. During the analysis, it was
found that the most effective factors of attraction are rain and night time (Table 2).
E-commerce recommender systems face challenges like the cold start problem, scalability, data
scarcity, and lack of contextual consideration. They tend to lack diversity, serendipity, and transparency
while struggling to adapt to dynamic user preferences and address privacy concerns. Advanced
recommender systems use hybrid approaches, context-aware recommendation, and deep learning
techniques to overcome these limitations.
ML and AI have greatly influenced recommender systems for e-commerce. Advancements in ML
and AI have had a significant impact on recommender systems in the field of e-commerce. Overall,
advancements in ML and AI have revolutionized recommender systems in e-commerce, enabling them
to deliver more accurate, personalized, and context-aware recommendations. Modern recommender

217
systems leverage techniques like deep learning and natural language processing (NLP) in various ways
to improve their recommendations. Some common approaches we are proposed the best known.
Table 2
Analysis of weather conditions
Weather conditions Variance
Snow -1.00%
Partially cloudy 2.90%
Rain 14.20%
Clear -3.60%
Overcast -12.40%

Collaborative filtering with deep learning – deep learning models, such as neural networks, can be
used to learn latent representations of users and items from historical user-item interactions. These
models can capture complex patterns and relationships in the data, enabling more accurate
recommendations. Techniques like matrix factorization and autoencoders are often employed for
collaborative filtering using deep learning.
Content-based filtering with NLP – NLP techniques are used to process and understand the textual
content associated with items or user preferences. Natural language processing algorithms can analyze
the text to extract features, such as keywords, topics, sentiment, or semantic meaning. These features
are then used to build item profiles or user profiles, enabling content-based filtering. NLP techniques
like text classification, named entity recognition, and sentiment analysis can be used for this purpose.
Neural embedding – deep learning models can learn dense vector representations, known as
embedding, for items and users. These embedding capture the underlying characteristics and
relationships between items and users. For example, an item embedding might encode information
about its genre, director, and actors. Similarly, a user embedding might represent their preferences and
behavior. These embedding can be learned using techniques like word2vec, GloVe, or more advanced
models like Transformers. Word2vec Word2Vec is a popular algorithm used in natural language
processing (NLP) to generate word embedding, which are vector representations of words in a high-
dimensional space. The word embedding produced by Word2Vec capture semantic and syntactic
relationships between words based on their context in a large corpus of text. The algorithm was
introduced by researchers Mikolov T. at Google in 2013. Word2Vec is a shallow neural network model
that learns word embedding by training on a large amount of text data. It makes use of a technique
called "distributed representations," where words with similar meanings or that often appear together
in context are represented by vectors that are close to each other in the vector space. There are two
primary architectures in Word2Vec: Continuous Bag-of-Words (CBOW) and Skip-gram.
There is the CBOW model. The CBOW model aims to predict a target word based on its surrounding
context words. It takes a sequence of context words as input and predicts the target word in the middle.
The model learns to associate the context words with the target word by adjusting the weights of the
neural network during training. There is the skip-gram model, on the other hand, takes a target word as
input and predicts the surrounding context words. It tries to maximize the probability of the context
words given the target word. The skip-gram model is generally slower to train but often produces better
word embedding, especially for infrequent words.
Both CBOW and skip-gram models learn word embedding by optimizing the neural network using
techniques such as stochastic gradient descent. The resulting word embedding can be used to measure
the similarity between words, perform analogical reasoning tasks, and enhance various NLP
applications such as text classification, information retrieval, and machine translation. Word2Vec has
been widely adopted due to its ability to capture semantic relation-ships between words and its
efficiency in training large-scale word embedding. It has played a significant role in advancing the field
of NLP and has paved the way for sub-sequent developments in word representation learning.
Innovation approaches and algorithms being used in state-of-the-art recommender systems for e-
commerce also are in particular graph-based models – graph-based approaches represent users, items,
and their relationships as a graph structure. Graph Convolutional Networks (GCNs) and Graph
Attention Networks (GATs) are used to capture the interdependencies between users and items,
enabling more accurate recommendations; session-based recommendations which focus on capturing
short-term user preferences based on their current session activity. Recurrent Neural Networks (RNNs),
Long Short-Term Memory (LSTM) networks, and Transformer-based models are used to model
sequential behavior and generate session-based recommendations, and knowledge graphs represent

218
items, users, and their attributes in a structured form. By incorporating knowledge graphs, recommender
systems can leverage semantic relationships between items and users to provide more contextually
relevant recommendations. It's important to note that the field of recommender systems is continuously
evolving, and researchers are actively working on developing new algorithms and techniques to
enhance recommendation quality and address various challenges.

3. Results
The Figure 2 represents the architecture of a system for an e-commerce platform, including various
components and their interactions. The system includes a “Customer” entity, an “API” package with
interfaces such as “OrderServiceAPI”, “ProductServiceAPI”, and “CartServiceAPI”, and a “Services”
package with components like “OrderServiceImpl”, “ProductServiceImpl”, and “CartServiceImpl”.
Additionally, there are the “Databases” packages with databases such as “OrderDatabase”,
“ProductDatabase”, and “CartDatabase”, and the “Message queues” package with queues named
“OrderQueue” and “CartQueue”.
Furthermore, the diagram showcases the “Recommender system” package consisting of the
following components: “RecommenderSystemAPI”: An interface facilitating interaction with the
recommendation system. It provides access to the functionality of the recommendation system for
communication with other components and external applications; “DataScienceService” - a service
responsible for data processing and scientific computations within the recommendation system. It
utilizes data from databases to analyze user preferences and offer personalized recommendations;
“Chatbot” - an interface for user interaction. Users can ask questions, and the chatbot uses the
“Knowledge_Base” to provide information and solutions. The “Knowledge_Base” contains
instructions, solutions, and other useful data that assist the chatbot in answering user questions and
providing recommendations; “Statistics_Records” - records statistical information on user requests. It
logs queries that the chatbot couldn't answer, enabling analysis of these situations and improvement of
the knowledge base.

Figure 2: E-store architecture with recommender system

The “Chatbot” component interacts as follows:
 The user asks the AI a question;
 The AI utilizes the knowledge base, which contains instructions on how to solve the issue;
 The AI responds to the customer, and if additional questions arise, it processes them as well;
 The AI records statistical information on user requests, primarily focusing on questions it
couldn't answer;

219
 The “Recommender system” package brings together components related to the recommendation
system, data processing, user interaction, and statistical analysis. It enables the provision of
personalized recommendations and system enhancement based on user interaction and statistics.

4. Conclusions
In light of the transformation into a digital society, the development of the field of electronic
economy, and in particular, electronic business and electronic commerce, is receiving a new impetus.
Therefore, consumer behavior models and recommender systems in e-commerce are of particular
importance. In the field of recommender systems for e-commerce, several future trends and directions
are emerging. These trends are as follows: personalization and contextualization, explainability and
transparency, usage of hybrid approaches, context-aware recommendations, group and social
recommendations, reinforcement learning, privacy and trust, multimodal recommendations, long-tail
recommendations, online learning and real-time recommendations, etc.
The focus is shifting towards highly personalized recommendations that consider individual
preferences, behaviors, and context. Advanced techniques such as deep learning, natural language
processing, and reinforcement learning are being employed to capture nuanced user preferences and
deliver tailored recommendations. There is a growing demand for explainable recommender systems.
Users want to understand why certain recommendations are made, particularly in the case of complex
algorithms like deep neural networks. Techniques like attention mechanisms, rule-based explanations,
and interpretable models are being explored to provide transparent recommendations.
Combining multiple recommendation techniques is gaining traction. Hybrid recommender systems
integrate collaborative filtering, content-based filtering, and knowledge-based approaches to leverage
the strengths of each method. This helps overcome limitations and provides more accurate and diverse
recommendations. Incorporating contextual information such as time, location, weather, and social
context can enhance the quality of recommendations. Context-aware recommender systems consider
temporal dynamics, user context, and situational factors to deliver more relevant and timely
suggestions. Recommending items to groups or communities rather than individuals is an evolving area.
Social recommender systems leverage social network information, user interactions, and group
dynamics to generate recommendations that align with collective interests and preferences.
Reinforcement learning techniques are being explored to optimize recommender systems. By using
feedback loops, recommender systems can actively learn and adapt to user feedback, continuously
improving the quality of recommendations over time. With increasing concerns about data privacy,
there is a focus on developing recommender systems that respect user privacy. Techniques like
federated learning, secure multi-party computation, and differential privacy are being researched to
provide privacy-preserving recommendations while maintaining data security.
With the rise of multimedia content, recommender systems are being adapted to handle multiple
modalities such as text, images, audio, and video. Techniques like multimodal fusion, deep learning,
and cross-modal retrieval are used to provide recommendations based on diverse types of data.
Traditional recommender systems tend to prioritize popular items, leading to limited exposure to niche
or long-tail items. Addressing this, efforts are being made to pro-vide recommendations that promote
serendipity, novelty, and diversity, ensuring users are exposed to a broader range of products.
Real-time recommendation systems that continuously adapt to evolving user preferences and
dynamics are gaining importance. Online learning techniques enable systems to update
recommendations in real-time, considering the most recent user interactions and changes in preferences.
These trends represent ongoing research and development efforts in the field of recommender systems
for e-commerce, aiming to provide more accurate, personalized, and context-aware recommendations
while addressing user concerns regarding transparency, privacy, and diversity. By understanding the
behavior and preferences of your customers, using recommender systems for e-commerce companies,
you can create better products, provide personalized experiences, and develop effective marketing
strategies that contribute to the development of e-commerce in general.

5. References
[1] M. Wenzl, Understanding Online Consumer Behaviors for a Better Customer Journey. Available
online: URL https://www.shipbob.com/blog/online-consumer-behavior/ (April 2021).

220
[2] J. Langer, S. Beckmann, How online shop owners can stimulate customers’ exploration of new
products. Journal of Retailing and Consumer Services, 2017, 34, pp. 32-42.
[3] C. Liao, J. Chen, Enhancing consumer engagement in e-commerce websites: An integrated model of
e-service quality, customer value, and switching costs. Journal of Retailing and Consumer Services,
2019, 50, pp. 23-32.
[4] E. Berghausen, C. Backhaus, When do consumers choose same-day versus next-day delivery in online
shopping? An investigation of the underlying decision-making processes. Journal of Retailing and
Consumer Services, 2020, 55, pp.102-109.
[5] T. Wagner, A. Benlian, T. Hess, The concept of digital affordances revisited: An exploratory study in
the context of business-to-consumer electronic commerce. Journal of Business Research, 2019, 98, pp.
365-376.
[6] L. Otto, D. Schniederjans, How should e-commerce platforms design their recommendation systems
to cope with multiple objectives? European Journal of Operational Research, 2017, 258(3), pp. 934-
947.
[7] I. Fedorko, R. Bacik, B. Gavurova, Technology acceptance model in e-commerce segment.
Management & Marketing, 13(4), pp. 1242-1256.
[8] Faradivah Dwi Azizah et al. Impulsive Buying Behavior: Implementation of IT on Technology
Acceptance Model on E-Commerce Purchase Decisions. Golden Ratio of Marketing and Applied
Psychology of Business, 2022, 2(1). doi:10.52970/grmapb.v2i1.173.
[9] Se Hun, Lim, Sukho Lee, J. Dan, Is Online Consumers’ Impulsive Buying Beneficial for E-Commerce
Companies? An Empirical Investigation of Online Consumers’ Past Impulsive Buying Behaviors.
Information systems management, 2017, vol. 34(1), pp. 85-100.
[10] Daduk Merdika, Mansur, Erni Tisnawati Sule, Dwi Kartini, Yevis Marty Oesman, Aditya Halim
Perdana Kusuma Putra, Nurul Chamidah. Moderating of the Role of Technology Theory to the
Existence of Consumer Behavior on e-commerce. Journal of Distribution Science, 2019, 17-7, pp. 15-
25.
[11] Hoda Ghavamipoor, S. Alireza Hashemi Golpayegani, Maryam Shahpasand, A QoS-sensitive model
for e-commerce customer behavior. Journal of Research in Interactive Marketing, 2017, 11(4), pp. 380-
397.
[12] K. Wakil, F. Alyari, M. Ghasvari, Z. Lesani, L. Rajabion, A new model for assessing the role of
customer behavior history, product classification, and prices on the success of the recommender
systems in e-commerce. Kybernetes, 2020, 49(5), pp. 1325-1346.
[13] Quan Zhang, Jian Yang, Model Construction of Hierarchical Polarization Characteristics Combined
with Social E-Commerce Consumer Behavior. Security and Communication Networks, 2022. doi:
10.1155/2022/9666677.
[14] E. Sina, Charandabi, Omidreza Ghanadiof. Evaluation of Online Markets Considering Trust and
Resilience: A Framework for Predicting Customer Behavior in E-Commerce. Journal of Business and
Management Studies. doi: 10.32996/jbms.
[15] Yu-Teng Jang, Pei-Shan Hsieh, Understanding consumer behavior in the multimedia context:
incorporating gamification in VR-enhanced web system for tourism e-commerce. Multimedia Tools
and Applications, 2021, 80, pp. 29339-29365.
[16] Mehdi Khosrow-Pour, D.B.A. Consumer Behavior, Organizational Development, and Electronic
Commerce: Emerging Issues for Advancing Modern Socioeconomies. 2009. doi: 10.4018/978-1-
60566-126-1.
[17] V. L. Pleskach, T. G. Zatonatska, Electronic commerce: textbook. Kyiv: Knowledge. 2007. 535 p.
[18] Research Online – Purchase Offline concept. Available online: URL
https://www.linkedin.com/pulse/customer-experience-per-unit-area-new-indicator-sales-julia-medvid
[19] COVID-19: Tracking the impact on media and consumer behavior. Available online: URL
https://www.nielsen.com/insights/2020/covid-19-tracking-the-impact-on-fmcg-and-retail/
[20] Wen Chao, The Impact of Quality on Customer Behavioral Intentions Based on the Consumer Decision
Making Process as Applied in E-Commerce. Doctor of Philosophy (Management Science), 2012, 144
pp.
[21] V. Zosimov, O. Bulgakova, Calculation the Measure of Expert Opinions Consistency Based on Social
Profile Using Inductive Algorithms. Advances in Intelligent Systems and Computing, 2020, 1020, pp.
622–636.
[22] V. Pleskach, V. Zosimov, O. Bulgakova, O. Nagornyi, Web-Based Management System for Customer
Interaction in E-Trade with Adaptive Interface. CEUR Workshop Proceedings, 2022, 3288, pp. 82-88.

221