Uncertainty-aware Generative Recommendation

TL;DR

This paper introduces UGR, a novel framework for generative recommendation that incorporates uncertainty to improve stability, performance, and confidence expression, addressing limitations of existing methods.

Contribution

UGR is the first unified framework to integrate uncertainty modeling into generative recommendation, enhancing stability and enabling risk-aware applications.

Findings

UGR outperforms existing methods in recommendation accuracy.

UGR stabilizes training dynamics and prevents performance degradation.

Learned confidence enables reliable downstream risk-aware tasks.

Abstract

Generative Recommendation has emerged as a transformative paradigm, reformulating recommendation as an end-to-end autoregressive sequence generation task. Despite its promise, existing preference optimization methods typically rely on binary outcome correctness, suffering from a systemic limitation we term uncertainty blindness. This issue manifests in the neglect of the model's intrinsic generation confidence, the variation in sample learning difficulty, and the lack of explicit confidence expression, directly leading to unstable training dynamics and unquantifiable decision risks. In this paper, we propose Uncertainty-aware Generative Recommendation (UGR), a unified framework that leverages uncertainty as a critical signal for adaptive optimization. UGR synergizes three mechanisms: (1) an uncertainty-weighted reward to penalize confident errors; (2) difficulty-aware optimization dynamics to prevent premature convergence; and (3) explicit confidence alignment to empower the model with confidence expression capabilities. Extensive experiments demonstrate that UGR not only yields superior recommendation performance but also fundamentally stabilizes training, preventing the performance degradation often observed in standard methods. Furthermore, the learned confidence enables reliable downstream risk-aware applications.