Impact of Users' Cultural Background on Multi-faceted Trust-based Recommender Systems

Abstract

Trust-based recommender systems usually overlook the cultural background of people when making suggestions. In this paper, we propose some strategies to include the home country of users in trust-based recommendation algorithms and we aim to understand if this information can improve the recommender system performance.

Keywords

• eol>Multi-faceted Reputation Model
• Trust-based Recommender Systems
• Social Relations
• Cultural Background of Users
• Web searching and information discovery
• Recommender systems

-

1. Introduction and when it ignores social relations. However, as that model overlooks the home country of the people providBuilding on the theories of homophily [ 1 ] and social influ- ing trust evidence, it does not support the investigation of ence [ 2 ], which associate social links with user similarity, the possible impact of the origin of people on consumer trust-based recommender systems leverage information feedback. about the trust between the members of a social network In [13, 14], the authors showed that cultural diferences to face the cold start problem afecting recommender between the people who provide reviews about items (in systems [ 3 ], especially the collaborative ones [ 4 ]. Trust that case, Eastern and Western social network members can be inferred by exploiting diferent types of informa- who rate restaurants) strongly impact the evaluation of tion, such as the occurrence of collaboration events that items. Thus, there is not only an individual perspective involve users [ 5 ] (such as bookmarking), the presence on the evaluation of products and services but also a more of friend relations in social networks [ 6, 7, 8 ], users’ rep- general influence given by the origin of the people who utation [ 9 ], but also the expression of support to other experience items and rate them. In this perspective, we statements, such as the setting of "like" or "useful" opin- propose an analysis of trust in recommender systems that ions on the reviews posted in the social network [ 10 ]. takes the home country of users into account for rating

In [ 11, 12 ], the authors presented a family of multi- prediction. Our idea is that of using the information about faceted trust-based recommender systems that employ the cities where people live as a proxy for their cultural diferent sources of public information about users’ rep- background and validate the impact on the performance utation, and trust between users, to improve collabora- of a multi-faceted trust-based recommender system. tive filtering results. They proposed the Multi-faceted Trust Model to define trust among users in a compositional, configurable way. That model flexibly integrates 2. Proposal social links with public anonymous feedback received by user profiles and user contributions in social networks.

The ofline experiments carried out in that work provided encouraging results, outperforming state-of-theart trust-based, and collaborative recommender systems in accuracy, ranking of items, and error minimization both when it uses complete information about the trust also their home country. For instance, if we aim at suggesting restaurants, two users coming from the same country have more chances to have similar tastes rather than two people coming from diferent continents. 2. We plan to extend the work described in [17] to compute the helpfulness of the reviews and use that information as evidence of trust. Indeed way of writing a review and consequently the review’s helpfulness could be connected to the home country of the users. For example, Kim et al. [18] found out that customers are likely to perceive the reviews from users of the same country as more helpful, regardless of valence or the number of reviews. Therefore, the helpfulness of reviews is moderated by reviewers’ and readers’ cultural backgrounds. This suggests that diferences in cultural background can be an element that afects the helpfulness of a review. 3. We plan to jointly exploit the origin country of users, and the item aspects they care about, to improve the computation of users’ similarity. We will extract item aspects from reviews and connect them to the home country of the users considering that people from a specific country could be more interested in some aspects than others.

Our proposal is to build a user model that contains the aspects that users cite more frequently in their reviews to describe their priorities. Subsequently, when computing the similarity between users, we could give more weight to the aspects that are in common between users given their home country. 4. We are interested in understanding to what extent the home country of users impacts other users’ personality traits. Specifically, the home country of users could be connected to other traits such as the Need for Cognition [19] or the Curiosity trait [20] of users and this could impact the rating prediction of the recommender system.

We will investigate the above points by means of an experiment on an extended version of the Multi-faceted Trust Model developed in [ 12, 11 ].

References

[1] M. McPherson , L. Smith-Lovin , J. M. Cook , Birds of a feather: Homophily in social networks , Annual Review of Sociology 27 ( 2001 ) 415 - 444 . URL: https://doi.org/10.1146/annurev. soc.27.1.415. doi:10.1146/annurev.soc.27.1 .415.

[2] P. Marsden , N. Friedkin , Network studies of social influence , Sociological Methods & Research 22 ( 1993 ) 127 - 151 . URL: https://doi. org/10.1177/0049124193022001006. doi: 10 .1177/ 0049124193022001006.

[3] F. Ricci , L. Rokach , B. Shapira , Recommender Systems: Techniques, Applications, and Challenges, Springer US , New York, NY, 2022 , pp. 1 - 35 . URL: https://doi.org/10.1007/978-1- 0716 -2197- 4 _1. doi: 10 .1007/978-1- 0716 -2197- 4 _ 1 .

[4] C. Desrosiers , G. Karypis , A comprehensive survey of neighborhood-based recommendation methods , in: F. Ricci , L. Rokach , B. Shapira , P. B. Kantor (Eds.), Recommender Systems Handbook , Springer US, Boston, MA, 2011 , pp. 107 - 144 . URL: https://doi.org/10.1007/978-0- 387 -85820- 3 _4. doi: 10 .1007/978-0- 387 -85820-3\_4.

[5] K. Mcnally , M. P. O'Mahony , B. Smyth , A comparative study of collaboration-based reputation models for social recommender systems, User Modeling and User-Adapted Interaction 24 ( 2014 ) 219 - 260 . URL: http://dx.doi.org/10.1007/s11257-013-9143-6. doi: 10 .1007/s11257-013-9143-6.

[6] M. Jamali , M. Ester , A matrix factorization technique with trust propagation for recommendation in social networks , in: Proceedings of the Fourth ACM Conference on Recommender Systems, RecSys '10 , ACM , New York, NY, USA, 2010 , pp. 135 - 142 . URL: http://doi.acm. org/10 .1145/1864708. 1864736. doi: 10 .1145/1864708.1864736.

[7] G. Guo , J. Zhang , N. Yorke-Smith, TrustSVD: Collaborative filtering with both the explicit and implicit influence of user trust and of item ratings , in: Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence , AAAI' 15 , AAAI Press, 2015 , pp. 123 - 129 . URL: http://dl.acm.org/citation.cfm? id= 2887007 . 2887025 .

[8] B. Yang , Y. Lei , J. Liu, W. Li , Social collaborative ifltering by trust , IEEE Transactions on Pattern Analysis and Machine Intelligence 39 ( 2017 ) 1633 - 1647 . doi: 10 .1109/TPAMI. 2016 . 2605085 .

[9] J. Tang , X. Hu , H. Gao , H. Liu, Exploiting local and global social context for recommendation , in: Proceedings of the Twenty-Third International Joint Conference on Artificial Intelligence, IJCAI '13 , AAAI Press, 2013 , pp. 2712 - 2718 . URL: http://dl. acm.org/citation.cfm?id= 2540128 .2540519. doi: 10 . 5555/2540128.2540519.

[10] G.-N. Hu , X.-Y. Dai , F.-Y. Qiu , R. Xia , T. Li , S.-J. Huang , J.-J. Chen , Collaborative filtering with topic and social latent factors incorporating implicit feedback , ACM Transactions on Knowledge Discovery from Data 12 ( 2018 ) 23 : 1 - 23 : 30 . URL: http://doi.acm. org/10 .1145/3127873. doi: 10 .1145/3127873.

[11] L. Ardissono , N. Mauro , A compositional model of multi-faceted trust for personalized item recommendation , Expert Systems with Applications 140 ( 2020 ) 112880 . URL: http://www.sciencedirect.com/