Understanding and Counteracting Feature-Level Bias in Click-Through Rate Prediction

Jinqiu Jin, Sihao Ding, Wenjie Wang, Fuli Feng

Abstract: Common click-through rate (CTR) prediction recommender models tend to exhibit feature-level bias, which leads to unfair recommendations among item groups and inaccurate recommendations for users. While existing methods address this issue by adjusting the learning of CTR models, such as through additional optimization objectives, they fail to consider how the bias is caused within these models. In this paper, we discover a generation path of feature-level bias: biased positive sample ratios → biased linear weights in CTR model → biased prediction scores → biased recommendations. Based on this understanding, we propose a minimally invasive yet effective strategy to counteract feature-level bias in CTR models by removing the biased linear weights from well-trained models. Additionally, we present a linear weight adjusting strategy that requires fewer random exposure records than relevant debiasing methods. The superiority of our proposed strategies are validated through extensive experiments on three real-world datasets. The code is available at https://github.com/mitao-cat/feature-level_bias