An enhanced probabilistic fairness-aware group recommendation by incorporating social activeness

Highlights

• Considering the fairness in group recommendation setting.

• Applying Coalition game theory to model social fairness.

• Considering social activeness for classifying groups.

Abstract

Compared with individual recommendation, recommending services to a group of users is more complicated because of various users' preference should be considered and introduces new challenging such as fairness, which has never been well studied in current works. In this paper, we propose a novel recommendation scheme called PFGR, which combines a probabilistic model with coalition game strategy, to ensure the accuracy and fairness between groups of users. Given a group of users and a set of services, PFGR models a generative process for service selection in light of several observations: 1) each group is related with several topics; 2) users' decisions on the service selection depends on their expertise, the opinions of members they are familiar with, and group influence; 3) each group contains active users and inactive user, whose activeness contributes to the existence of group. PFGR first estimates the preference of each user on a candidate service via combining user's expertise, inherent connection, and group influence. Then, it determines a group's decision on a service by aggregating the preference of group members using adaptive weights. Finally, PFGR considers users' activeness and employs a strategy based on coalition game to produce a ranked list which is fair to each group member as much as possible. Experimental results on three real-world datasets validate that PFGR can achieve higher Hit Rate and Average Reciprocal Hit Rank than state-of-the-art approaches, which indicates that PFGR attains both the precision and fairness of recommendation.

Introduction

Traditional recommender systems (RSs) aim to provide appropriate services for a single user based on her preferences. Such RSs have been deployed in a wide range of areas such as music (Yahoo), restaurants (Foursquare), and hiking (Meetup). However, many contexts requires recommending to a group of users (i.e., group recommendation) while various preferences of all the group members should be considered. For example, in cases of selecting a picnic location for a group of friends, recommending a restaurant for a company's annual meeting, arranging attractions for a group of tourists, the traditional individual recommendation methods no longer fit.

Group recommendation is more complicated than individual recommendation. Since group members may have different preferences (Yuan et al., 2014; Carvalho and Macedo, 2013a), a service preferred by one user may not satisfy another user's taste. Moreover, each user hopes her preferred service to appear at a top position in the service list recommended to her group. According to the studies in the fair division of sources (Manurangsi and Suksompong, 2017; Aleksandrov et al., 2015; Herreiner and Puppe, 2009), a recommended services list is fair to a user if and only if her preferred service is ranked at a top position (Serbos et al., 2017). Therefore, it is of paramount importance to recommend a ranked service list that is fair to every user, i.e., fairness. An ideal recommendation approach for group not only guarantee the accuracy but also efficiently solve fairness issue.

Most current studies on group recommendation (Baltrunas et al., 2010; Liu et al., 2012; Rakesh et al., 2016; Yuan et al., 2014; Quijano-Sanchez et al., 2013; Salehi-Abari and Boutilier, 2014; Ronen et al., 2014) determines the services that satisfy the group members' preferences via modelling users' implicit peer influence (Yuan et al., 2014). However, they cannot solve the fairness issue because they commonly lack a proper method to balance the various preferences. Other studies (Serbos et al., 2017; Xiao et al., 2017; Zhang et al., 2017; Carvalho and Macedo, 2013a, 2013b) convert the fairness issue into a comparison sequencing problem and design a preference-based sequencing strategy to rank the recommended services. Although this strategy can ensure fairness to some extent, it cannot tackle the scenarios where group members have conflicting preferences. As it is intractable to compare users' preference (e.g., distinguish the optimal options from spicy and light food preferences), the recommended list derived by this strategy can only guarantees a part of users' preferences instead of all the users’ preferences. Therefore, sequencing strategy based on preference is improper.

Fortunately, the social regularization principle (O'Hara, 1999) provides a interesting viewpoint: the more contribution you pay, the more priority or return you win (Marx, 2008). For a group, its formation and sustainability heavily depends on its members' activeness, which refers to as the frequency of users' interactions including sharing information or extending the social circle (Turner, 1982). Inspired by the social regularization principle, it is more intuitive and proper to consider users' activeness when ranking services, i.e., a user's preference should be satisfied in priority if she contributes to the group more actively. Different from dealing with users' preferences, we can easily quantify users' activeness via simple statistic methods (Koller et al., 2007) and handle conflicting user preferences. For example, we can count up how many friends a user has or how much shopping information she shares.

We borrow the fairness definition from (Serbos et al., 2017; Xiao et al., 2017) and propose a novel two-stage group recommendation model called PFGR. PFGR couples user's various preferences and activeness, which has seldom been studied by previous work. PFGR consists of two parts: multi-facet probabilistic graph model (MFPG) and activeness-based coalition game strategy (ACG). During recommendation, PFGR first applies MFPG to produce the services which satisfy all the members' preferences by modeling several observations (see Section 3.5) obtained from the real life. Then, it utilizes ACG to rank these services to attain a trade-off among various preferences.

Specifically, MFPG is a probabilistic generative model that aims to select the services preferred by a group. It is developed on latent Dirichlet allocation (LDA), which has been proven successful in modeling implicit interactions (Blei et al., 2003; Jelodar et al., 2017). Compared with other group recommendation model based on LDA (Yuan et al., 2014; Rakesh et al., 2016), MFGP considers more implicit interactions such as users' social links, preferences influence, and common-interest. In particular, considering users' implicit interactions can help group members to better select their desired services. ACG is inspired by the coalition game theory, which has two advantages when compared with current sequencing strategy based on the greedy algorithm (Serbos et al., 2017; Xiao et al., 2017) or the non-cooperative game theory (Zhang et al., 2017; Carvalho and Macedo, 2013b): 1) instead of considering a single user's preference, the coaliton game theory innately considers users' peer influence (e.g., common-interest, social links) and therefore conforms to the fact that a user's selection may be affected by others; 2) the coalition game theory considers the balance between several coalitions. That makes it easier to find the equilibrium among a large number of users in a dynamic environment where each user's preferences may change over time.

We make the following contributions in this paper:

• We propose a novel two-stage group recommendation approach named PFGR which both guarantee the accuracy of recommendation and efficiently solve fairness issue. PFGR couples users' preferences and activeness, which has not been well studied before.

• we design an activeness-based sequencing strategy to ranking services following the social regularization principle to promote the fairness in recommendation. This strategy can better solve conflicted preference contexts when compared with the traditional preference-based sequencing strategy.

• We conduct extensive experiments to validate the effectiveness of PFGR under various settings on three real data sets. The evaluation results show our scheme consistently outperforms state-of-the-art approaches when considering the fairness simultaneously.

The rest of this paper is organized as follows: Section 2 reviews the related work. Section 3 introduces the preliminaries and formulates the group recommendation problem. Section 4 presents the details of our proposal, including the MFPG model and ACG strategy. Section 5 reports our analysis of experiment results. Finally, Section 6 concludes the paper.