Utility optimization-based multi-stakeholder personalized recommendation system
| DOI | https://doi.org/10.1108/DTA-07-2021-0182 |
| Published date | 15 April 2022 |
| Date | 15 April 2022 |
| Pages | 782-805 |
| Subject Matter | Library & information science,Librarianship/library management,Library technology,Information behaviour & retrieval,Metadata,Information & knowledge management,Information & communications technology,Internet |
| Author | Rahul Shrivastava,Dilip Singh Sisodia,Naresh Kumar Nagwani |
Utility optimization-based
multi-stakeholder personalized
recommendation system
Rahul Shrivastava, Dilip Singh Sisodia and Naresh Kumar Nagwani
Department of Computer Science and Engineering,
National Institute of Technology Raipur, Raipur, India
Abstract
Purpose –In a multi-stakeholde r recommender system (M SRS), stakeholders ar e the multiple entitie s
(consumer, producer, system, etc.) benefit ed by the generated recommend ations. Traditionally, the exclusive
focus on only a single stak eholders’(for example, only consumer or end-user) preferences obscured the
welfare of the others. Two major challe nges are encountered while inco rporating the multiple stakeho lders’
perspectives in MSRS: de signing a dedicated util ity function for each st akeholder and optimiz ing
their utility without hurting others. This paper proposes multiple utilit y functions for different
stakeholders and opt imizes these functio ns for generating bal anced, personalized recommendations for
each stakeholder.
Design/methodology/approach –The proposed methodology considers four valid stakeholders user,
producer, cast and recommender system from the multi-stakeholder recommender setting and builds
dedicated utility functions. The utility function for users incorporates enhanced side-information-
based similarity computation for utility count. Similarly, to improve the utility gain, the authors
design new utility functions for producer, star-cast and system to incorporate long-tail and diverse
items in the recommendation list. Next, to balance the utility gain and generate the trade-off
recommendation solution, the authors perform the evolutionary optimization of the conflicting utility
functions using NSGA-II. Experimental evaluation and comparison are conducted over three
benchmark data sets.
Findings –The authors observed 19 .70% of average enhance ment in utility gain wit h improved
mean precision, diver sity and novelty. Expos ure, hit, reach and targe t reach metrics are sub stantially
improved.
Originality/value –A new approach considers four stakeholders simultaneously with their respective utility
functions and establishes the trade-off recommendation solution between conflicting utilities of the
stakeholders.
Keywords Multi-stakeholder recommender system, Diverse, Long-tail, Utility
Paper type Research paper
1. Introduction
End-user personalization in the recommendation system (RS) is an important criterion for
managing overwhelming information content in any application area, such as e-commerce
(Aggarwal, 2016;Pujahari and Sisodia, 2020a). The application areas like e-commerce, end-
users are not the only stakeholder in the system (Burke et al., 2016). For example, apart from
movie users in a movie recommendation, other stakeholders, including producer, star-cast
and system itself, are considered valid stakeholders (Pujahari and Sisodia, 2019),
(Abdollahpouri, 2019;Pujahari and Sisodia, 2022). Reciprocal recommendations such as
job recommendation (Mine et al., 2013) dating applications (Pizzato et al., 2010;Xia et al., 2015)
is the real-world application of multi-stakeholder recommender system (MSRS). Multi-sided
fairness in MSRS may be achieved by incorporating the preferences of each stakeholder in the
RS (Evans et al., 2011).
MSRS recently adopted a utility function-oriented approach. For example, educational-domain
based MSRS (Zheng et al., 2019) develop utility functions for the three stakeholders: student,
DTA
56,5
782
The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/2514-9288.htm
Received 20 July 2021
Revised 20 October 2021
1 March 2022
Accepted 28 March 2022
Data Technologies and
Applications
Vol. 56 No. 5, 2022
pp. 782-805
© Emerald Publishing Limited
2514-9288
DOI 10.1108/DTA-07-2021-0182
teacher and system. The utility optimization-based model proposed in (Zheng and Pu, 2019)
generates the trade-off between conflicting utilities of stakeholders. The utility
functions in these methods obtain utility for each stakeholder from the top-n
recommendation list (RL). Popular personalized top-nrecommendation techniques
(Hernando et al.,2016;Luo et al., 2012;Pujahari and Sisodia, 2020b) and textual content-
basedRS(Shrivastava and Sisodia, 2019) mainly focus on improving recommendation
accuracy. The utility functions targeting only end-user preferences may achieve better
accuracy but lacks novel and diverse item recommendations (Ge et al.,2010;Vargasand
Castells, 2011).
This study investigates the problem of enhancing top-nrecommendation quality and
utility gain of stakeholders by exploiting the utility function-oriented approach. In this
work, each stakeholder’s dedicated utility function is designed to obtain personalized
preference-based utilities. The stakeholders’utilities are conflicting in nature. Therefore,
we further propose a stakeholder utility optimization-based recommendation framework
(SURF) to balance and optimize conflicting utilities and enhance the utility gain of all
stakeholders. We consider a multi-stakeholder recommendation scenario over a movielens
(ML) and book-crossing data set as a case study to demonstrate the working of SURF. We
consider four stakeholders: the user, producer/publisher, star-cast/author and
recommender system.
To the best of our knowledge, no study till date personalizes the recommendation
for four stakeholders in a single execution. Following are the main contributions of
this study:
(1) For the first time, this study considers four stakeholders simultaneously and
develops the personalized utility functions for each stakeholder.
(2) This study embeds beyond accuracy objectives in utility functions to produce diverse
and long-tail item recommendations.
(3) The integrated framework SURF is proposed to balance the stakeholders’conflicting
utilities by employing evolutionary optimization and improving personalization for
stakeholders.
(4) The proposed framework is evaluated across three benchmark data sets utilizing
multi-stakeholder utility gain and exclusive evaluation measures for each
stakeholder, including precision, diversity, novelty, exposure, hit, reach and
target reach.
The rest of the articleis organized as follows: Section2 presents the literaturereview; Section 3
describes the proposed utility function-based recommendation scenario. Further, Section 4
shows experimental results, Section 5 presents result analysis and discussion and Section 6
concludes the paper.
2. Related work and motivation
The MSRS incorporates beyond user preferences, and similarly, the multi-objective
recommendation model establishes a trade-off between beyond accuracy preferences
of the user (Cui et al., 2017;Jain et al., 2020). The novelty (Vargas and Castells, 2011),
serendipity and diversity (Jungkyu and Yamana, 2017), based non-accuracy objectives
may be incorporated in the MSRS to enhance the stakeholders’utility gain.
Evolutionary multi-objective optimization models (Cui et al., 2017;Jain et al., 2020)
have been developed to enhance the economic strength of e-commerce applications by
providing a trade-off recommendation solution between precision, diversity and
Personalized
recommendation
system
783
To continue reading
Request your trial