Unpaired Modality-Agnostic Generative Recommendation

2026-08-03

AI Summary

This work addresses the limitations of existing multimodal generative recommendation methods, which rely on paired data and are constrained by modality-specific semantic ID spaces, thereby struggling to effectively leverage unpaired unimodal observations. To overcome this, the authors propose a unified semantic ID space framework that enables modality-agnostic generative recommendation through a shared Transformer encoder, a residual vector quantization codebook, and lightweight modality-specific projection heads. This design seamlessly integrates paired, image-only, and text-only data without requiring feature imputation or multiple codebooks. Extensive experiments on three benchmark datasets demonstrate that the proposed method significantly outperforms current state-of-the-art models under both complete and missing modality settings, confirming its strong cross-modal generalization capability and practical utility.

Abstract

Generative Recommendation (GR) formulates recommendation as autoregressive generation over discrete semantic identifiers (IDs). Although recent multimodal GR methods improve semantic ID construction with visual and textual information, they typically require item-level paired observations, restricting tokenization to the intersection of modality availability. Moreover, incorporating unpaired observations is nontrivial because small representation shifts may cross quantization boundaries and produce incompatible identifier sequences. To address this challenge, we propose \textbf{Unpair}ed Modality-Agnostic \textbf{G}enerative \textbf{R}ecommendation (UnpairGR), which learns a unified semantic-ID space from paired, image-only, and text-only observations. UnpairGR confines modality-specific processing to lightweight input projections while sharing the subsequent Transformer and residual codebooks across all observation conditions. Paired observations establish a reliability-guided cross-modal consensus, whereas unimodal observations directly refine the same representations and codes. The learned tokenizer is then fixed to provide stationary targets for a single autoregressive recommender, without feature imputation, modality-specific codebooks, or fallback mappings. Extensive experiments on three benchmark datasets demonstrate that UnpairGR consistently improves recommendation performance under both fully observed and incomplete-observation settings.

Problem

• Generative Recommendation
• Unpaired Multimodal Data
• Semantic Tokenization
• Modality-Agnostic Learning
• Quantization Boundaries

Innovation

• Generative Recommendation
• Modality-Agnostic
• Unpaired Multimodal Learning
• Semantic Tokenization
• Unified Representation