Multi-Agent Personalized Recommendation

System in E-Commerce based on User

Venkateswara Reddy B
Dr. Nagagopiraju Vullam
Professor, Dept. of CSE,
Clustering
Sai Srinivas Vellela
Assistant Professor, Dept. of CSE,
Assistant Professor, Dept. of CSE,
Chalapathi Institute of Technology,
Chalapathi Institute of Chalapathi Institute of Technology,
Guntur, AP, India
Technology, Guntur, AP, India Guntur, AP, India
[email protected]
[email protected] [email protected]
systemmis made to help customers find what they want and
complete their purchases successfully. E-commerce
Roja D
Dr. M Venkateswara Rao Khader Basha recommendation
Sk systems help customers complete the
Assistant Professor, Dept. of CSE,
Professor, Dept. of CSE, Assistant Professor,purchase process by providing them with product
2023 2nd International Conference on Applied Artificial Intelligence and Computing (ICAAIC) | 978-1-6654-5630-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICAAIC56838.2023.10140756

Dept. of CSE, Chalapathi Institute of Technology,

Chalapathi Institute of Technology, information
Chalapathi Institute of Technology,and suggestions, including suggestions for
Guntur, AP, India
Guntur, AP, India Guntur, AP,products
India that may be of interest to them. These systems
[email protected] m
[email protected] [email protected]

Abstract: As more sectors began to switch from are based on an understanding of consumers interests and
conventional business models to e-commerce in response to preferences. Network users have a possibility of
the general trend toward mobile Internet use, the scale of e - learning more information because of the flood of
commerce grew rapidly. There are three types of different types of information caused by the expansion of
recommendation systems: hybrid, collaborative, content- network data. However, users of the network have a
based. Content based systems take into consideration the significant need to select useful information from
characteristics of the recommended objects. Then, titles in
numerous information sources. Because e-commerce
the database that have been classified as "romantic" are
selected using a content-based recommendation
transaction partners do not actually interact, it is both a
method.Collaborative filtering systems utilize similarity new opportunity and a new problem for businesses to
measures to recommend items that are shared by individuals implement every element of the previous business model
or objects with similar interests. Users are recommended through the Internet.
items based on their preferences.In the recommendation
system, collaborative filtering is the most popular and As a result of the widespread use of mobile Internet, an E-
effective suggestion process. However, system performance commerce is growing quickly in China as more
impact as the amount of time required to locate the target businesses start to replace traditional methods of busines s
user's closest neighbor across the entire user space increases with e-commerce.As per to their own investment
with the number of users and products in the e -commerce characteristics and purchasing patterns, the
system. The applied and designed Multi-Agent personalized recommendation system can analyse users' needs and
recommendation system in E-commerce can be analyzed
interests. It can then search through large amounts of
using user clustering in the Multi -Agent to E-commerce
information to identify and recommend to users the
personalized recommendation system. An implementation
strategy for recommendations based on user clusteri ng is content that matches their interests[1].
shown in this analysis. According to their scores for
Due to the rapid expansion of the Internet and mobile
commodity categories, users are clustered, and only the
nearest neighbours in their categories are searched, so that
Internet, the number of e-commerce websites is
as many nearest neighbors as possible can be searched. The constantly growing, as well as the rise in Chinese internet
accuracy, recall, and specificity of this analysis are used to users, online consumers, and their participation rates[2].
calculate its performance. In this analysis the presented Customers can receive completely personalised
method will give better results. information and decision support services from an
intelligent platform known as a personalised
KEYWORDS :User clustering; Multi-Agent;
Personalization, E-commerce; Recommendation system.
recommendation system that is based on big data mining.
The recommendation system's primary goals are to
I.INTRODUCTION provide users with suggestions, assist users in finding the
required information and permit information providers to
By communicating with clients directly and simulating offer a number of content to genuine users who are
shop salespeople while providing product interested in it in order to meet their specific
recommendations, An e-commerce recommendation requirements. In addition to being an important future
development direction, the provision of personalized
service is an important issue that requires urgent attention
in order to enhance the quality of network content
services. From the artificial intelligence (AI) field the
concept of Agent It depicts a piece of is
originated.software that can operate continuously and
autonomously, solve problems and fix them, and
communicate with the outside world:

1194
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.
Intelligent: With comparable accuracy, the agent can customised suggestion and non-personalized
speculate the user's intent. It provides the information you recommendation are the two basic types of
require, analyses complex processes, highlights problems, recommendation systems.DingZhaoqiaoet. al.,
and simplifies them. As circumstances change, it can
provide customers with specialised services based on their [4] Demographic recommendations can’t be utilised as a
particular needs. recommendation technology in recommendation systems,
which can’t only solve the scarcity problem. Even though
Reactivity: Agents are capable of understanding their they use different data sets for advice, this method and
environment and react quickly to relevant objects. collaborative filtering recommendation technologies are
improvement and learning for themselves: Because Agent essentially similar. because the item's scoring matrix
is a independent entity, it may work with other users contains a high number of missing values and the u ser
while simultaneously only scored a small percentage of the items. This makes
the score prediction challenge one of how to fill in the
developing new abilities independently. It might convert missing numbers using the present scoring values.
directly into an internal model of the user's interests,
activities, and other data, store it in a knowledge base, and [6]Yan Yinget. al., In contrast to the recommendation
create a user model to help it make decisions. system, the recommendation system actively recommends
relevant content that can actually meet users requirements
Sociality: An agent operates in coordination in a multi- and interests by analyzing various user attributes and user
agent system and digests conflicts through cooperation history records.
and communication. Some difficult tasks that are
impossible to complete on their own can be completed [9]S. Varma, D. Sharma M. Jain, and A. Beniwalet.
with this assistance. In order to understand the potential al.Some well-known commercial search engines, like
needs of the customers, the agent is proactive in gathering Yahoo!, utilize a technique called query suggestion, Bing,
and reviewing their personal data. Based on the user's Ask, and Google, to help users find relevant queries and
preferences, it enables Agent to offer services proactively. make internet browsing simple. The query
Agent is capable of operating autonomously. Some of its recommendation feature can predict an active user's
own actions and internal factors are under its control. interests by looking at data from similar users or products.
The query recommendation feature severely limits the
A system with multiple Agents is referred to as a Multi- scope of the search.
Agent System (MAS). The key to logically coordinating
several Agents with different objectives and maximising [10]H. Ma, I. King and M. R. Lyuet. al.User-based
the success of each Agent's target is to combine many recommendations work by recommending items liked by
Agents with these goals. There are different kinds of other users, item-based recommendations highlight
agents, each with a different goal, knowledge, and products based on similar features. Collaborative filtering
capability. A network of agents that work together to has recently been used several matrix factorization
solve problems is being created. A multi-agent system is approaches. These methods integrate the user-item rating
better than a one-agent system because it can improve matrix and produce predictions by using low-rank
each agent's essential abilities and solve problems that are approximations.
too difficult for one agent to handle. These Agent and
Multi-Agent properties suggest that using Agent to [11]P. W. Yau and A. Tomlinson et. al.,In content-based
develop a customised recommendation system contains filtering, recommendations are made based on the
its advantages.A user clustering-based implementation qualities of the products or services. A product or
technique for cooperative suggestions is shown in this service's qualities are compared to those of other
analysis. Users are classified into commodity categories properties in the database after being analyzed. Then,
based on their scores, and in as little user space as goods with comparable qualities are suggested. On e-
possible, only those users nearest neighbors. commerce websites, content-based filtering is frequently
used. Content-based filtering was successfully employed
Remaining analysis is arranged as follows: by websites like Amazon, eBay, SnapDel, and Flipkart to
propose products to users.
literature survey can be explain in Section II, In this
Section III describes a Multi-Agent personalised [12]J. Jiang, J. Lu, G. Zhang and G. Long et. al., On an e-
Recommendation System in E-commerce that is based on commerce website, when looking for a specific product, a
user clustering., the results analysis is describes in section user can see a list of questions that other users have
IV and the conclusion is described in section V. already searchedOn websites like Amazon, Flipkart,
Snapdeal, and others, query recommendations are utilised
to enhance the user's browsing experience.
II. LITERATURE SURVEY [13]Ming Zhu, Jun Wang and Junpu Wang et. al.,
[3]N. Karbhari, A. Deshmukh, and V. D. Shindeet. al. A Personal assistance for clients at their visited websites
recommendation system is a method for gathering which is known as personalised recommendations in E-
information with the goal of developing an algorithm that commerce.E-commerce recommendation system
takes into account the various needs and levels of categorises customers based on factors like browsing
competence when providing recommendations. The

1195
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.
frequency and length. Within a cluster, browsed-page sets
User Browse
are highly similar to one another and to other clusters.

Web server
III. MULTI-AGENT PERSONALIZED
RECOMMENDATION SYSTEM IN E-COMMERCE
BASED ON USER CLUSTERING
Search engine
Figure 1 shows the block diagram of the presented Multi-
Agent personalized Recommendation System in E-
commerce based on user clustering. Recommendation engine Agent
Personalized e-commercial recommendation system was
intended to construct as a model for online
recommendations on business websites. During the client
session, search engines are used to generate dynamic page
ideas based on the user's current viewing. Its primary Evaluation User Profile Data mining
purpose is to suggest web pages. A search engine is Feedback Management Agent
essential for the online component because it is primarily Agent Agent
responsible for calculating and defining the page that is
set for the current user and is designed to his interests. At
first the users current session is analyzed by the search
engine. It obtains a list of the websites the user has
already visited after establishing their current session. The Recommendation Algorithm
user's request is forwarded by the search engine to the
web server. Search engines send the recommended page
to users when new pages are returned from Web
servers.This completes a page recommendation.

The system's heart is the Agent recommendation engine. Product Evaluation

It was created as a MultiAgent process is presented using Information Records
a search engine, a recommendation generator, and a Library Library
recommendation control agent. Figure 1 provides a
detailed explanation of its structure. The user profile
management agent, the Evaluation Feedback Agent
communicates with the user interface agent, the user
surveillance agent, the data mining agent, and between User Library Recommen
the user agents.
dation
Records
Database

Data Mining

Library

Fig. 1: THE BLOCK DIAGRAM OF PRESENTED

MULTI-AGENT PERSONALIZED
RECOMMENDATION SYSTEM IN E-COMMERCE
BASED ON USER CLUSTERING

The evaluation input agent receives the user assessment

results from the user interface agent, evaluation result is
stored in the user evaluation library. Utilizing a data
collection agent, it learns and predicts user feedback
before updating the user profile dynamically.This could
help Agents, the recommendation engine, detect the user's
interests more precisely and improve the precision of
recommendations.Online and offline evaluation are the
two techniques used to assess recommendations. They

1196
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.
both work together in the assessment feedback Agent the data is subsequently stored in a library of evaluation
designs. They can considerably increase the accuracy of records for self-improvement and future usage.
users' resource evaluations and assist in the collection of
more precise and authentic data on user preferences and IV. RESULT ANALYSIS
interests. The practise of assessing online survey results is The result analysis of presented Multi-Agent personalized
known as online assessment. The User Interface Agent's Recommendation System in E-commerce based on user
user feedback box is for online evaluation. The evaluation clustering is demonstrated in this section.
feedback agent receives the evaluation results. The performance of the developed model is evaluated
The user agent, evaluation feedback agent, and data using the following definitions for False Negative (FN),
mining agent are all in communication with the user False Positive (FP) True Positive (TP), and True Negative
profile management agent. Its primary duty is to create (TN):
and maintain user interest models. An individualized True Positive (TP): The number of accurately identified
service is built on the foundation of the creation of a user factually positive predictive cases that are all positive
interest model. Getting user interest is the first step in predictions is known as the total probabilities (TP).
creating a user interest model, and the second is updating
user data and interest.Collecting User Interests: client True Negative (TN): The total number of correctly
demographics User agents and management agents categorised negative predictive cases is known as the TN.
interact directly to exchange information about
False Positives (FP)are instances of positive prediction
requirements and preferences. Similarly, assessing, data
outcomes that are mistakenly classified as such and are
mining and monitoring agents may indirectly identify user
not factually positive.
interests. The latter is more proactive and particularly
suited to user-specific recommendations systems. A False Negative (FN): The total number of instances of
crucial first step in identifying the interests of present false negative predictive modelling that are classified as
consumers is exploring consumer demands and feedback. incorrect is known as FN.
Data mining Agent : The primary responsibility of data
mining agents, who get requests from recommendation Accuracy: The ratio of accurately detected occurrences to
engine agents is to investigate client purchasing and all instances is provided, and it is referred to as
browsing activities using methods including decision
trees, association rules, Web log mining,neural networks,
regression algorithms. Data mining Agent gathers user
demands and interests in this manner to support user Recall: Recall shows a dataset's capacity to contain every
profile management Agent with tasks. Additionally, it relevant instance. The result is expressed as:
simultaneously develops association rules for client Classification models identify all relevant events in a
classification, assessment prediction from similar recollection
customer base and customer access to products. All of
them will be the foundation blocks on which Agent's (2)
recommendation engine will be constructed.
Using content-based recommendation algorithms, Specificity:It is written as and defined as the ratio of
consumers may widely regarded for the same or similar actual negative cases to true negative occurrences (FP +
products based on past buying habits or interests. The TN)
content-based recommendation system, examines the
similarities among films that users have viewed and (3)
received higher ratings before recommending other
The table 1 describes the performance analysis of
movies to users that have scored higher.This is true in the
presented Multi-Agent personalized Recommendation
case of movie recommendations.Offline Evaluation: It
System in E-commerce based on user clustering.
actually uses data mining to determine potential
assessment based on the present evaluation by using user Table 1: PERFORMANCE ANALYSIS
suggestion records and profile page libraries. By
conducting these evaluations into collaborative filtering, it Performance Mahout System M-APRS In E-
is possible to overcome the sparse nature of collaborative Metrics Dynamics Commerce
filtering and increase the concentration of resource based on user
clustering
evaluations, both of which improve collaborative
filtering's efficiency and raise customer satisfaction with Accuracy (%) 89 85.3 98.8
the suggested outputs.
The monitoring agent directs the Recommendation Recall 90.1 86 97.6
Generation Agent to choose recommendations based on (%)
product collection, databases, user interest algorithms, Specificity 88.4 85.1 97.2
and information (the number of products evaluated by (%)
users and the number of users in existing user models).
The recommendation record library briefly stores the
The above table shows that thepresented Multi-Agent
created concepts for review and agent feedback.User
personalized Recommendation System in E-commerce
satisfaction is calculated using assessment findings, and

1197
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.
based on user clustering gives the highest accuracy, V. CONCLUS ION
recall and specificity.
Customers might get from Personalized recommendations
in e-commerce when they visit website. Personalized
suggestion services have become hot spot as e-commerce
has developed rapidly. This analysis introduces user
clustering and multi-Agent technology to create a
specialized, highly intelligent, and personalized
recommendation system. The user clustering technology
used in this study differs from collaborative filtering
recommendation technology in that it analyzes user
interests based on objective features rather than the
subjective preferences of the user. Accuracy, recall, and
specificity are different aspects used to assess this
method's performance. Compared to the Multi-Agent
Customized Recommendation System in E-Commerce,
which is based on user clustering, these results are better.

Fig. 2: THE ACCURACY COMPARISION REFERENCES

BETWEEN THE METHODS [1] Zhang Wen, Li Jian, Cui Yangbo, et al. Clustering matrix
approximation-based collaborative filtering recommendation research
In this comparison the above graph shows that Multi- Volume of Operations Research and Management 169, no. 4, pp. 175 -
Agent personalized Recommendation System in E- 182, 2020.
commerce based on user clustering has a higher level of [2] Zhou Zhiyuan. Predict user behavior using data from e-commerce
and machine learning. Volume of Communication World 26, no. 3, pp.
accuracy compared to the Mahout and system dynamics -
212-213, 2019.
based recommendation systems. [3] N. Karbhari, V. D. Shinde, and A. Deshmukh, “Content -filtering
recommendation system: A case study for college campus placement, in
ICECDS 2017, the 2017 International Conference on Energy,
Communication, Data Analytics, and Soft Computing.
[4] Sk, K. B., Vellela, S. S., Yakubreddy, K., & Rao, M. V. (2023).
Novel and Secure Protocol for T rusted Wireless Ad-hoc Network
Creation. Khader Basha Sk, Venkateswara Reddy B, Sai Srinivas
Vellela, Kancharakunt Yakub Reddy, M Venkateswara Rao, Novel and
Secure Protocol for Trusted Wireless Ad-hoc Network Creation, 10(3).
[5] K. B. Sk, R. D, S. S. Priya, L. Dalavi, S. S. Vellela and V. R. B,
"Coronary Heart Disease Prediction and Classification using Hybrid
Machine Learning Algorithms," 2023 International Conference on
Innovative Data Communication T echnologies and Application
(ICIDCA), Uttarakhand, India, 2023, pp. 1-7, doi:
10.1109/ICIDCA56705.2023.10099579.
[6] Yan Ying. Study of a Cluster-Based Collaborative Filtering
Algorithm. Changsha Telecommunications Vocational and T echnical
College Journal, vol. 16, no. 4, pp. 40-43, 2017.
[7] VenkateswaraRao, M., Vellela, S., Reddy, V., Vullam, N., Sk, K. B.,
& Roja, D. (2023, March). Credit Investigation and Comprehensive Risk
Fig. 3: THE RECALL COMPARIS ION BETWEEN Management System based Big Data Analytics in Commercial Banking.
THE METHODS In 2023 9th International Conference on Advanced Computing and
Communication Systems (ICACCS) (Vol. 1, pp. 2387-2391). IEEE.
[8] HuiZheng, Bing Li, Donglin Chen, et al. Based on location clusters,
personalized recommendation technology for mobile life services.
Volume on Computer Applications 35, no. 4, pp. 1148 -1153, 2015.
[9] M Venkateswara Rao,Sai Srinivas Vellela, Khader Basha Sk,
Venkateswara Reddy B, Roja D, 6Lavanya Dalavai(2023),
SYST EMAT IC REVIEW ON SOFT WARE APPLICAT ION
UNDERDIST RIBUT ED DENIAL OF SERVICE AT T ACKS FOR
GROUP WEBSIT ES, Dogo Rangsang Research Journal UGC Care
Group I Journal, Vol-13, Issue-3, No. 6, ISSN : 2347-7180, March 2023
[10] Vellela, S. S., &Balamanigandan, R. (2022, December). Design of
Hybrid Authentication Protocol for High Secure Applications in Cloud
Environments. In 2022 International Conference on Automation,
Computing and Renewable Systems (ICACRS) (pp. 408-414). IEEE.
[11] Venkateswara Reddy B , Sai Srinivas Vellela , Khader Basha Sk,
Roja D , Kancharakunt Yakubreddy , M Venkateswara Rao, Conceptual
Hierarchies for Efficient Query Results Navigation, International
Journal of All Research Education and Scientific Methods (IJARESM),
ISSN: 2455-6211 Volume 11, Issue 3, March-2023.
[12] Vellela, S. S., Balamanigandan, R., & Praveen, S. P. (2022).
Strategic Survey on Security and Privacy Methods of Cloud Computing
Fig. 4: THE SPECIFICITY COMPARISION Environment. Journal of Next Generation T echnology (ISSN: 2583 -
BETWEEN THE METHODS 021X), 2(1).

1198
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.
[13] P. W. Yau and A. T omlinson, "Towards Privacy in a ContextAware
Social Network Based Recommendation System," in Privacy, Security,
Risk, and T rust (PASSAT ) and the 2011 IEEE T hird International
Conference on Social Computing (SocialCom), Boston, MA, 2011,
pages 862-865.
doi:10.1109/PASSAT /SocialCom.2011.87
[14] Vellela, S. S., & Krishna, A. M. (2020). On Board Artificial
Intelligence With Service Aggregation for Edge Computing in Industrial
Applications. Journal of Critical Reviews, 7(07), 2020.
[15] Madhuri, A., Jyothi, V. E., Praveen, S. P., Sindhura, S., Srinivas, V.
S., & Kumar, D. L. S. (2022). A New Multi-Level Semi-Supervised
Learning Approach for Network Intrusion Detection System Based on
the ‘GOA’. Journal of Interconnection Networks, 2143047.
[16] J. Jiang, J. Zhang Lu, G. and G. Long, "Increasing T hing Based
Cooperative Sifting Proposal Calculation In light of Hadoop," 2011
IEEE World Congress on Administrations, Washington,
DC,2011,pp.490-497. doi: 10.1109/SERVICES.2011.66
[17] Vellela, S. S., Reddy, B. V., Chaitanya, K. K., &Rao, M. V. (2023,
January). An Integrated Approach to Improve E-Healthcare System
using Dynamic Cloud Computing Platform. In 2023 5th International
Conference on Smart Systems and Inventive Technology (ICSSIT ) (pp.
776-782).IEEE.
[18] Madhuri, A., Praveen, S. P., Kumar, D. L. S., Sindhura, S.,
&Vellela, S. S. (2021). Challenges and issues of data analytics in
emerging scenarios for big data, cloud and image mining. Annals of the
Romanian Society for Cell Biology, 412-423.
[19] Yakubreddy, K., Vellela, S. S., Sk, K. B., Reddy, V., &Roja, D.
GRAPE CS-ML DAT ABASE-INFORMED MET HODS FOR
CONT EMPORARY VINEYARD MANAGEMENT .
[20] Babu, G. B., Gopal, M. V., Srinivas, V. S., & Krishna, V. Efficient
Key Generation for Multicast Groups Based on Secret Sharing.
[21] Ming Zhu, Junpu Wang and Jun Wang. T he Investigation of
Feature Selection in the Classification of Web Pages[J]. 26 (8),
Computer Engineering, 2000: 35-37.
[22] Praveen, S. P., Sarala, P., Kumar, T. K. M., Manuri, S. G., Srinivas,
V. S., &Swapna, D. (2022, November). An Adaptive Load Balancing
T echnique for Multi SDN Controllers.In 2022 International Conference
on Augmented Intelligence and Sustainable Systems (ICAISS) (pp.
1403-1409).IEEE.
[23] SaiSrinivasVellela , KhaderBashaSk, Venkateswara Reddy B, Roja
D, KancharakuntYakubreddy “Cloud-hosted concept -hierarchy flex-
based infringement checking system” International Advanced Research
Journal inScience, Engineering and T echnology ISO 3297:2007
Certified, ISSN (O) 2393-8021, ISSN (P) 2394-1588,Impact Factor
7.12, Vol. 10, Issue 3, March 2023 DOI:
10.17148/IARJSET .2023.10308.
[24] V. R. B, K. BashaSk and R. D, "Qos-Aware Video Streaming based
Admission Control and Scheduling for Video T ranscoding in Cloud
Computing," 2022 International Conference on Automation, Computing
and Renewable Systems (ICACRS), Pudukkottai, India, 2022, pp. 520 -
525, doi: 10.1109/ICACRS55517.2022.10029194.
[25] Karthik, J. V., & Reddy, B. V. (2014). Authentication of secret
information in image stenography. International Journal of Computer
Science and Network Security (IJCSNS), 14(6), 58.

1199
Authorized licensed use limited to: STATE UNIV NY BINGHAMTON. Downloaded on June 11,2023 at 06:59:36 UTC from IEEE Xplore. Restrictions apply.