Test Time Embedding Normalization for Popularity Bias Mitigation

TL;DR

This paper introduces 'Test Time Embedding Normalization', a simple method that significantly reduces popularity bias in recommender systems by normalizing item embeddings during inference, improving fairness without sacrificing accuracy.

Contribution

The paper presents a novel test-time normalization technique for item embeddings that effectively mitigates popularity bias in recommendation models.

Findings

Reduces popularity bias more effectively than previous methods

Normalizing embeddings during inference controls bias without harming recommendation quality

Angular similarity between user and item embeddings correlates with preference regardless of popularity

Abstract

Popularity bias is a widespread problem in the field of recommender systems, where popular items tend to dominate recommendation results. In this work, we propose 'Test Time Embedding Normalization' as a simple yet effective strategy for mitigating popularity bias, which surpasses the performance of the previous mitigation approaches by a significant margin. Our approach utilizes the normalized item embedding during the inference stage to control the influence of embedding magnitude, which is highly correlated with item popularity. Through extensive experiments, we show that our method combined with the sampled softmax loss effectively reduces popularity bias compare to previous approaches for bias mitigation. We further investigate the relationship between user and item embeddings and find that the angular similarity between embeddings distinguishes preferable and non-preferable items regardless of their popularity. The analysis explains the mechanism behind the success of our approach in eliminating the impact of popularity bias. Our code is available at https://github.com/ml-postech/TTEN.