Contrastive Counterfactual Learning for Causality-Aware Interpretable Recommender Systems

By Guanglin Zhou et al.

Read the original document by opening this link in a new tab.

Table of Contents

ABSTRACT
KEYWORDS
1 INTRODUCTION
2 PRELIMINARY
3 THE PROPOSED METHOD
3.1 Causality-Aware Interpretation of Recommendations
3.2 Exposure Bias Reduction
3.3 Contrastive Counterfactual Learning
3.3.1 Contrastive Counterfactual Learning (CCL)
3.3.2 Three Novel Positive Samplings

Summary

The document presents a novel approach, Contrastive Counterfactual Learning, for addressing exposure bias in recommender systems. It emphasizes the importance of considering causal interpretation in recommendations and proposes methods to mitigate exposure bias through contrastive self-supervised learning. The proposed Contrastive Counterfactual Learning (CCL) method incorporates three novel positive sampling strategies grounded in counterfactual inference. By maximizing agreements between anchor and positive samples, CCL aims to reduce exposure bias and improve the interpretability and fairness of recommender systems.