RASTP: Representation-Aware Semantic Token Pruning for Generative Recommendation with Semantic Identifiers

TL;DR

RASTP is a method that improves generative recommendation systems by dynamically pruning less informative semantic tokens, reducing computational costs while maintaining or enhancing recommendation accuracy.

Contribution

It introduces a novel token pruning technique based on semantic saliency and attention centrality, directly reducing input sequence length in generative recommendation models.

Findings

Reduces training time by 26.7% on Amazon datasets

Maintains or slightly improves recommendation performance

Demonstrates effectiveness of token pruning in real-world datasets

Abstract

Generative recommendation systems typically leverage Semantic Identifiers (SIDs), which represent each item as a sequence of tokens that encode semantic information. However, representing item ID with multiple SIDs significantly increases input sequence length, which is a major determinant of computational complexity and memory consumption. While existing efforts primarily focus on optimizing attention computation and KV cache, we propose RASTP (Representation-Aware Semantic Token Pruning), which directly prunes less informative tokens in the input sequence. Specifically, RASTP evaluates token importance by combining semantic saliency, measured via representation magnitude, and attention centrality, derived from cumulative attention weights. Since RASTP dynamically prunes low-information or irrelevant semantic tokens, experiments on three real-world Amazon datasets show that RASTP reduces training time by 26.7\%, while maintaining or slightly improving recommendation performance. The code has been open-sourced at https://github.com/Yuzt-zju/RASTP.