Rethinking Semantic ID Construction for Generative Recommendation: SimHash with Parallel Decoding and Semantic Alignment

2026-10-05

AI Summary

This study addresses the challenge of constructing semantic IDs that balance expressiveness and efficiency in generative recommendation. We reveal that the performance bottleneck of hashing methods stems from a mismatch with autoregressive decoding architectures rather than their inherent limitations. Accordingly, we propose FLASH, a framework employing a training-free SimHash tokenizer integrated with a parallel decoding architecture and an explicit semantic alignment mechanism, enabling efficient, high-quality recommendation without tokenizer training. Experiments demonstrate that FLASH achieves state-of-the-art performance across multiple datasets while significantly enhancing cold-start generalization. These results confirm that simple hashing, when paired with an appropriate decoding mechanism, can rival complex learned quantization approaches, thereby validating the universal effectiveness of semantic alignment.

Abstract

Semantic ID-based generative recommendation represents each item as a sequence of discrete tokens, enabling structured modeling of item semantics. A critical challenge is constructing semantic IDs that are both semantically expressive and computationally efficient. While recent approaches favor complex learned quantization, simple hashing-based methods such as SimHash are widely regarded as fundamentally inferior. In this work, we challenge this consensus by showing that the apparent performance gap does not stem from inherent limitations of hashing, but rather from a structural mismatch with autoregressive decoding, coupled with the inevitable information loss during rigid discretization. Based on this insight, we propose FLASH, a two-stage framework that revitalizes training-free SimHash tokenization through parallel decoding and explicit semantic alignment. Despite its simplicity, FLASH achieves state-of-the-art performance across multiple datasets without requiring any tokenizer training, while exhibiting stronger generalization in cold-start scenarios. Notably, we demonstrate that semantic alignment acts as a universally effective mechanism across diverse paradigms. Our findings suggest that, with compatible decoding and semantic grounding, simple and efficient tokenizers can achieve performance comparable to complex learned counterparts in generative recommendation. Our code is available at https://github.com/KevinC2015/Flash.

Problem

• Generative Recommendation
• Semantic ID Construction
• SimHash
• Autoregressive Decoding
• Semantic Alignment

Innovation

• Generative Recommendation
• SimHash
• Parallel Decoding
• Semantic Alignment
• Semantic ID