Unbiased Collaborative Filtering with Fair Sampling

TL;DR

The paper addresses popularity bias in implicit-feedback recommender systems by showing that bias arises from propensity factors during training. It proposes Fair Sampling (FS), a propensity-free sampling framework with FS-Point and FS-Pair that augments training data to prevent propensity-factor amplification in both point-wise and pair-wise losses. The authors provide theoretical analysis demonstrating the elimination of propensity influence without estimating propensity scores, and empirically show state-of-the-art performance and reduced bias on three real-world datasets. Acknowledging a limitation that FS may overlook popularity signals, they suggest future work to jointly balance diversity and quality via sampling-ratio adjustments.

Abstract

Recommender systems leverage extensive user interaction data to model preferences; however, directly modeling these data may introduce biases that disproportionately favor popular items. In this paper, we demonstrate that popularity bias arises from the influence of propensity factors during training. Building on this insight, we propose a fair sampling (FS) method that ensures each user and each item has an equal likelihood of being selected as both positive and negative instances, thereby mitigating the influence of propensity factors. The proposed FS method does not require estimating propensity scores, thus avoiding the risk of failing to fully eliminate popularity bias caused by estimation inaccuracies. Comprehensive experiments demonstrate that the proposed FS method achieves state-of-the-art performance in both point-wise and pair-wise recommendation tasks. The code implementation is available at https://github.com/jhliu0807/Fair-Sampling.