FairSR: Fairness-aware Sequential Recommendation through Multi-Task Learning with Preference Graph Embeddings

Abstract

Sequential recommendation (SR) learns from the temporal dynamics of user-item interactions to predict the next ones. Fairness-aware recommendation mitigates a variety of algorithmic biases in the learning of user preferences. This article aims at bringing a marriage between SR and algorithmic fairness. We propose a novel fairness-aware sequential recommendation task, in which a new metric, interaction fairness , is defined to estimate how recommended items are fairly interacted by users with different protected attribute groups. We propose a multi-task learning-based deep end-to-end model, FairSR, which consists of two parts. One is to learn and distill personalized sequential features from the given user and her item sequence for SR. The other is fairness-aware preference graph embedding (FPGE). The aim of FPGE is two-fold: incorporating the knowledge of users’ and items’ attributes and their correlation into entity representations, and alleviating the unfair distributions of user attributes on items. Extensive experiments conducted on three datasets show FairSR can outperform state-of-the-art SR models in recommendation performance. In addition, the recommended items by FairSR also exhibit promising interaction fairness.

A Next Basket Recommendation Reality Check

“…Configurations. To assess group fairness (Section 4), we follow configurations from previous research [26,28]; the group of items is determined by their popularity (i.e., the number of purchases recorded in the historical baskets of the dataset). The top 20% of items with the highest purchase frequency as the popular group (𝐺 + ), while the remaining 80% of items are assigned to the unpopular group (𝐺 − ).…”

MDAR: A Knowledge-Graph-Enhanced Multi-Task Recommendation System Based on a DeepAFM and a Relation-Fused Multi-Gead Graph Attention Network

“…Some recent studies have focused on disseminating multi-layered relational information within the knowledge graph to enhance recommendations. For instance, researchers have employed relation-aware graph neural networks [18,[22][23][24][25][31][32][33][34][35][36] to capture neighborhood features. RippleNet [18] naturally integrates the KGE method into the recommendation system through the preference propagation method, thus continuously and automatically excavating the user's potential hierarchical interest.…”

“…CKAN [24] tackles the cold start problem by using knowledgeaware collaborative networks to improve recommendation accuracy. FairSR [25] applies a fairness-aware preference graph embedding method to incorporate the knowledge of users' and items' attributes and their correlation into entity representations and alleviate the unfair distributions of user attributes on items. DFM-GCN [31] and DeepFM_GCN [32] employ the DeepFM and GCNs to learn the entity representation in the RS and the KGE, to excavate the semantic feature information between the two modules and improve the accuracy of recommendation.…”

“…The precision rate is used to indicate how many of the items that the model predicts users may be interested in are items that the user will actually be interested in when the users have not interacted with items. The formula of precision is as follows in Equation (25).…”

“…Graph neural networks (GNNs) that aggregate node attributes from neighborhoods using neural networks are usually exploited to learn the representations in KGs in an efficient, explicit, and end-to-end manner. In this way, auxiliary information can be exploited to enrich the representations of users, entities, and relations, such as constructing user KGs [18][19][20][21], item KGs [22][23][24][25], or user-item KGs [26][27][28][29][30], and we can extract more accurate preferences of users by excavating the relationships between entities. In recent years, several researchers have applied factorization machines and neural networks to MKR, drawing on the strengths of both by alternate training, thus improving the performance of the recommendation systems.…”