Balancing User Preferences by Social Networks: A Condition-Guided Social Recommendation Model for Mitigating Popularity Bias

TL;DR

This work tackles popularity bias in social recommendation by jointly denoising social networks and guiding diffusion-based recommendations with adjusted social preferences. The proposed CGSoRec framework comprises a Condition-Guided Social Denoising Model (CSD) to remove redundant social relations and a Condition-Guided Diffusion Recommendation Model (CGD) to steer recommendations using refined social signals. Through weight-adjusted social preferences and conditional diffusion, the method mitigates hot-item over-recommendation while boosting long-tail item coverage, as demonstrated on three real-world datasets with thorough ablations. The results show that CGSoRec improves debiased ranking metrics and reduces popularity bias, highlighting the value of integrating denoising and conditional diffusion for fairer, more balanced recommendations. The work provides a practical, open-source approach for bias-aware social recommendation and sets a principled direction for future conditional diffusion-based debiasing research.

Abstract

Social recommendation models weave social interactions into their design to provide uniquely personalized recommendation results for users. However, social networks not only amplify the popularity bias in recommendation models, resulting in more frequent recommendation of hot items and fewer long-tail items, but also include a substantial amount of redundant information that is essentially meaningless for the model's performance. Existing social recommendation models fail to address the issues of popularity bias and the redundancy of social information, as they directly characterize social influence across the entire social network without making targeted adjustments. In this paper, we propose a Condition-Guided Social Recommendation Model (named CGSoRec) to mitigate the model's popularity bias by denoising the social network and adjusting the weights of user's social preferences. More specifically, CGSoRec first includes a Condition-Guided Social Denoising Model (CSD) to remove redundant social relations in the social network for capturing users' social preferences with items more precisely. Then, CGSoRec calculates users' social preferences based on denoised social network and adjusts the weights in users' social preferences to make them can counteract the popularity bias present in the recommendation model. At last, CGSoRec includes a Condition-Guided Diffusion Recommendation Model (CGD) to introduce the adjusted social preferences as conditions to control the recommendation results for a debiased direction. Comprehensive experiments on three real-world datasets demonstrate the effectiveness of our proposed method. The code is in: https://github.com/hexin5515/CGSoRec.

Figures (7)

• Figure 1: Popularity bias comparison between introduced and removed Social Network versions of DiffNet model. Items are categorized into six groups based on their popularity within the training set. The x-axis represents the range of frequencies at which each group of items was recommended. The left y-axis showing the proportion of each item group, and the right y-axis indicating the average frequency of recommendation.
• Figure 2: Average frequency of items recommended based on a user's social preferences. The x-axis represents the range of frequencies at which each group of items was recommended.
• Figure 3: The overall framework of the proposed method. We first derive the user-item interaction vector $\mathbf{x}$ and the user-user interaction matrices $\mathbf{s}^{cpl}$ and $\mathbf{s}$ based on different user-item and user-user relationships. Then we calculate the interaction vector $\mathbf{s}_{i}'$ to guide the social denoising model in the denoising process of the social interaction vector $\mathbf{s}_{i}$. Finally, we use the denoised user social interaction vector $\bar{\mathbf{s}}_{i}$ to calculate the user's social preferences. And we employ the weight adjusted preference $\mathbf{x}_{i}'$ to guide the diffusion recommendation model to obtain bias-reduced recommendation results $\bar{\mathbf{x}}_{i}$.
• Figure 4: The Joint Inferencing process of the proposed method. The interaction vector $\mathbf{s}_{i}'$ serves as a condition to guide each step of the denoising process in the condition-guided social denoising model, ultimately resulting in the denoised interaction vector $\bar{\mathbf{s}}_{i}$. Similarly, the weight adjusted user social preference vector $\mathbf{x}_{i}'$ also serves as a condition to guide each step of the denoising process in the condition-guided diffusion recommendation model, ultimately yielding bias-reduced recommendation results $\bar{\mathbf{x}}_{i}$.
• Figure 5: Frequency of different item groups recommended by DiffNet, LightGCN-S, MHCN and CGSoRec before and after the incorporation of social networks in LastFM and Ciao dataset.