Asymmetric Generative Recommendation via Multi-Expert Projection and Multi-Faceted Hierarchical Quantization

2026-05-14

AI Summary

This work addresses the dual information bottleneck in generative recommendation caused by symmetrically using discrete semantic IDs for both input and output, which leads to semantic loss, popularity bias, and imprecise supervision signals. To overcome these limitations, the authors propose AsymRec, a novel framework that introduces an asymmetric continuous–discrete architecture for the first time. Specifically, Multi-expert Semantic Projection (MSP) decouples input representations to preserve the rich semantics of continuous embeddings, while Multi-faceted Hierarchical Quantization (MHQ) constructs high-capacity, structured discrete output targets. A semantic regularization mechanism is further incorporated to mitigate dimensional collapse and enhance fine-grained discriminability. Extensive experiments demonstrate that AsymRec outperforms state-of-the-art generative recommendation models by an average of 15.8% across multiple benchmarks, significantly improving both recommendation accuracy and coverage of long-tail items.

Abstract

Generative Recommendation (GenRec) models reformulate recommendation as a sequence generation task, representing items as discrete Semantic IDs used symmetrically as both inputs and prediction targets. We identify a critical dual-stage information bottleneck in this design: (1) the Input Bottleneck, where lossy quantization degrades fine-grained semantics, while popularity bias skews the learned representations toward frequent items, and (2) the Output Bottleneck, where imprecise discrete targets limit supervision quality. To address these issues, we propose AsymRec, an asymmetric continuous-discrete framework that decouples input and output representations. Specifically, Multi-expert Semantic Projection (MSP) maps continuous embeddings into the Transformer's hidden space via expert-specialized projections, preserving semantic richness and improving generalization to infrequent items. Multi-faceted Hierarchical Quantization (MHQ) constructs high-capacity, structured discrete targets through multi-view and multi-level quantization with semantic regularization, preventing dimensional collapse while retaining fine-grained distinctions. Extensive experiments demonstrate that AsymRec consistently outperforms state-of-the-art generative recommenders by an average of 15.8 %. The code will be released.

Problem

• Generative Recommendation
• Information Bottleneck
• Semantic Quantization
• Popularity Bias
• Discrete Representation

Innovation

• Asymmetric Generative Recommendation
• Multi-Expert Semantic Projection
• Multi-Faceted Hierarchical Quantization
• Information Bottleneck
• Semantic ID