Repeat-bias-aware Optimization of Beyond-accuracy Metrics for Next Basket Recommendation

2025-01-10

AI Summary

To address the excessive preference for repeated items in basket recommendation—leading to degraded diversity and item fairness—this paper proposes RDA-Rerank, a model-agnostic post-processing optimization framework termed Repetition Bias-Aware Reranking. It is the first work to systematically characterize how repetition bias negatively impacts beyond-accuracy metrics (e.g., diversity and fairness), explicitly modeling and mitigating such bias via gradient reweighting, constrained optimization, and reranking. The framework is modular, supports multiple variants, integrates seamlessly with mainstream neural basket recommendation (NBR) models, and adapts to diverse data distributions. Evaluated on three real-world grocery datasets, RDA-Rerank improves diversity by 12.7%, item fairness by 18.3%, and reduces repetition bias by 31.5%, while incurring a negligible recall drop of less than 1.2%.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

In next basket recommendation (NBR) a set of items is recommended to users based on their historical basket sequences. In many domains, the recommended baskets consist of both repeat items and explore items. Some state-of-the-art NBR methods are heavily biased to recommend repeat items so as to maximize utility. The evaluation and optimization of beyond-accuracy objectives for NBR, such as item fairness and diversity, has attracted increasing attention. How can such beyond-accuracy objectives be pursued in the presence of heavy repeat bias? We find that only optimizing diversity or item fairness without considering repeat bias may cause NBR algorithms to recommend more repeat items. To solve this problem, we propose a model-agnostic repeat-bias-aware optimization algorithm to post-process the recommended results obtained from NBR methods with the objective of mitigating repeat bias when optimizing diversity or item fairness. We consider multiple variations of our optimization algorithm to cater to multiple NBR methods. Experiments on three real-world grocery shopping datasets show that the proposed algorithms can effectively improve diversity and item fairness, and mitigate repeat bias at acceptable Recall loss.

Problem

• Fairness
• Diversity
• User Preference

Innovation

• Intelligent Algorithm
• Diversity and Fairness
• Recommendation Optimization