Leave No One Behind: Enhancing Diversity While Maintaining Accuracy in Social Recommendation

TL;DR

The paper addresses the tendency of social recommendation to enhance accuracy at the expense of diversity. It introduces DivSR, a model-agnostic framework that employs relational knowledge distillation to transfer high-diversity knowledge from a non-social teacher model to a social student model, using an angle-based diversity loss $\ l_D$ and balancing it with the primary ranking loss $\mathcal{L}_R$ via $\mathcal{L}=\mathcal{L}_R(\theta_s)+\beta\mathcal{L}_D(\theta_s|\theta_t)$. The key contributions include an empirical study revealing the accuracy-diversity tension in existing social RS, the design of a diversity-knowledge transfer mechanism, and extensive experiments on Yelp, Ciao, and Flickr across multiple backbones showing DivSR improves diversity with minimal accuracy loss. The work demonstrates that structured knowledge transfer can produce superior accuracy-diversity trade-offs and offers a practical, reusable approach for deploying diverse yet accurate social recommendations. The public release of code and data facilitates reproducibility and adoption in real-world systems.

Abstract

Social recommendation, which incorporates social connections into recommender systems, has proven effective in improving recommendation accuracy. However, beyond accuracy, diversity is also crucial for enhancing user engagement. Despite its importance, the impact of social recommendation models on diversity remains largely unexplored. In this study, we systematically examine the dual performance of existing social recommendation algorithms in terms of both accuracy and diversity. Our empirical analysis reveals a concerning trend: while social recommendation models enhance accuracy, they often reduce diversity. To address this issue, we propose Diversified Social Recommendation (DivSR), a novel approach that employs relational knowledge distillation to transfer high-diversity structured knowledge from non-social recommendation models to social recommendation models. DivSR is a lightweight, model-agnostic framework that seamlessly integrates with existing social recommendation architectures. Experiments on three benchmark datasets demonstrate that DivSR significantly enhances diversity while maintaining competitive accuracy, achieving a superior accuracy-diversity trade-off. Our code and data are publicly available at: https://github.com/ll0ruc/DivSR.

Figures (5)

• Figure 1: The trends of the user-friend embedding similarity during training.
• Figure 2: The overall framework of DivSR.
• Figure 3: Accuracy-Diversity trade-off comparison on Yelp dataset.
• Figure 4: User embeddings visualization.
• Figure 5: Parameter sensitivity.