Challenging Low Homophily in Social Recommendation

TL;DR

This work proposes Social Heterophily-alleviating Rewiring (SHaRe), a data-centric framework for enhancing existing graph-based social recommendation models and integrates a contrastive learning method into the training of SHaRe, aiming to calibrate the user representations for enhancing the result of Graph Rewiring.

Abstract

Social relations are leveraged to tackle the sparsity issue of user-item interaction data in recommendation under the assumption of social homophily. However, social recommendation paradigms predominantly focus on homophily based on user preferences. While social information can enhance recommendations, its alignment with user preferences is not guaranteed, thereby posing the risk of introducing informational redundancy. We empirically discover that social graphs in real recommendation data exhibit low preference-aware homophily, which limits the effect of social recommendation models. To comprehensively extract preference-aware homophily information latent in the social graph, we propose Social Heterophily-alleviating Rewiring (SHaRe), a data-centric framework for enhancing existing graph-based social recommendation models. We adopt Graph Rewiring technique to capture and add highly homophilic social relations, and cut low homophilic (or heterophilic) relations. To better refine the user representations from reliable social relations, we integrate a contrastive learning method into the training of SHaRe, aiming to calibrate the user representations for enhancing the result of Graph Rewiring. Experiments on real-world datasets show that the proposed framework not only exhibits enhanced performances across varying homophily ratios but also improves the performance of existing state-of-the-art (SOTA) social recommendation models.

Figures (11)

• Figure 1: Preference-aware homophily ratio distributions of social graphs on three real-world datasets, where $h_{(i,j)}$ is the edge-wise homophily ratio of user-user edge $(u_i, u_j)$, $\mathcal{H}_s$ is the graph-wise homophily ratio of the social graph (see definitions in Section \ref{['sec:problem_def']}).
• Figure 2: The influence of graph-wise homophily ratio to different social recommendation models.
• Figure 3: An overview of the proposed SHaRe framework. A recommendation encoder learns user embeddings $P$ from the interaction graph. These user embeddings $P$ are used for rewiring the social relations matrix $\bm{S}$. The rewired social relations matrix $\widehat{\bm{S}}$ and the interaction matrix $\bm{R}$ are inputted to the backbone social recommendation models. Their output user and item representations are used for calculating the recommendation loss $\mathcal{L}_{\text{rec}}$.
• Figure 4: User embedding similarity and Social Graph Rewiring (SGR).
• Figure 5: SHaRe-DiffNet results under different graph-wise homophily ratios.