Long-Tail Session-based Recommendation from Calibration

TL;DR

This paper introduces a calibration-based approach to session-based recommendation systems that balances popularity bias by aligning tail item ratios with user preferences, improving diversity without sacrificing accuracy.

Contribution

It proposes a novel calibration module and a two-stage curriculum training strategy to better reflect user preferences for long-tail items in recommendations.

Findings

Achieves competitive recommendation accuracy

Provides more tail items in recommendations

Effectively mitigates popularity bias

Abstract

Accurate predictions in session-based recommendations have progressed, but a few studies have focused on skewed recommendation lists caused by popularity bias. Existing models for mitigating popularity bias have attempted to reduce the overconcentration of popular items by amplifying scores of less popular items. However, they normally ignore the users' different preferences toward long-tail items. Thus, we incorporate calibration, where calibrated recommendations reflect the users' interests in recommendation lists with appropriate proportions, to mitigate the popularity bias from the user's perspective. Specifically, we propose a calibration module to predict the ratio of tail items in the recommendation list from the session representation, and align it to the ongoing session. Additionally, we utilize a two-stage curriculum training strategy to improve prediction in the calibration module. Experiments on benchmark datasets show that our model can both achieve the competitive accuracy of recommendation and provide more tail items.