HCGRec: Hint-Conditioned Generative Recommendation with Semantic IDs

2026-08-12

AI Summary

This work addresses the challenge in reward-based post-training for generative recommender systems with semantic ID representations, where early token errors often lead to a high proportion of zero-reward samples, hindering effective optimization. To mitigate this issue, the authors propose a prompt-conditioned generation framework that employs a target-prefix prompting mechanism to provide minimal guidance when generation deviates, steering the model back to the correct semantic trajectory. Additionally, they introduce a prompt-aware credit assignment strategy to recover meaningful learning signals from hard examples. By integrating semantic ID representations, checkpoint-based rollout diagnostics, and a hybrid optimization approach based on GRPO, the method substantially outperforms both supervised fine-tuning and standard reward-based post-training on sequential recommendation benchmarks, reducing the fraction of zero-advantage samples from over 70% to below 20%.

Abstract

Semantic-ID generative recommenders represent each item as a short sequence of discrete semantic tokens and predict the next item by autoregressively generating this token sequence. This paradigm enables a unified generation interface for item IDs, histories, and item text, but it also creates a structured optimization bottleneck during reward-based post-training: when an early semantic token enters the wrong branch of the item-token space, finite rollout groups rarely reach the ground-truth item, so group-relative optimization receives identical zero rewards and produces no useful advantage. We propose Hint-Conditioned Generative Recommendation (HCGRec), a semantic-ID generative recommendation framework that recovers learning signal for such hard training instances. HCGRec diagnoses each instance with checkpoint rollouts and supplies a minimal target-prefix hint only when the current generator cannot reach the correct item. The model then generates the unhinted suffix under the hinted semantic branch, turning zero-reward groups into informative comparisons over item-token completions. Hinting also changes token identity: hinted prefix tokens are oracle-provided item context, while unhinted suffix tokens are sampled generation actions. We therefore introduce hint-aware credit decomposition, using supervised learning to preserve item-semantic and prefix-structure alignment for hinted tokens and GRPO to optimize the sampled suffix. Experiments on sequential recommendation benchmarks show that HCGRec substantially improves over supervised fine-tuning and vanilla reward-based post-training, while reducing zero-advantage training samples from over 70% to below 20%. The code is accessible at https://github.com/WncFht/GRec.

Problem

• generative recommendation
• semantic IDs
• reward-based post-training
• optimization bottleneck
• zero-reward problem

Innovation

• Hint-Conditioned Generation
• Semantic ID
• Reward-Based Post-Training
• Credit Decomposition
• Generative Recommendation