Learnable Item Tokenization for Generative Recommendation

2024-05-12

International Conference on Information and Knowledge Management

AI Summary

This work addresses three key challenges in LLM-driven generative recommendation: coarse-grained semantic representation of items, absence of collaborative signals, and biased codebook allocation in item-to-language-space mapping. To this end, we propose LETTER—the first end-to-end learnable item tokenizer for generative recommendation. LETTER jointly models hierarchical semantics, user-item collaborative relationships, and codebook diversity within a residual-quantized variational autoencoder (RQ-VAE) framework. It incorporates contrastive alignment loss to bridge modality gaps, diversity regularization to ensure balanced codebook usage, and ranking-guided generation loss to directly optimize recommendation performance. Extensive experiments on three public benchmarks demonstrate that LETTER consistently outperforms state-of-the-art baselines, achieving the first SOTA results for LLM-based generative recommendation. Our approach establishes a new paradigm for deep integration of recommender systems and large language models.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Utilizing powerful Large Language Models (LLMs) for generative recommendation has attracted much attention. Nevertheless, a crucial challenge is transforming recommendation data into the language space of LLMs through effective item tokenization. Current approaches, such as ID, textual, and codebook-based identifiers, exhibit shortcomings in encoding semantic information, incorporating collaborative signals, or handling code assignment bias. To address these limitations, we propose LETTER (a LEarnable Tokenizer for generaTivE Recommendation), which integrates hierarchical semantics, collaborative signals, and code assignment diversity to satisfy the essential requirements of identifiers. LETTER incorporates Residual Quantized VAE for semantic regularization, a contrastive alignment loss for collaborative regularization, and a diversity loss to mitigate code assignment bias. We instantiate LETTER on two models and propose a ranking-guided generation loss to augment their ranking ability theoretically. Experiments on three datasets validate the superiority of LETTER, advancing the state-of-the-art in the field of LLM-based generative recommendation.

Problem

• Transform recommendation data into LLM language space
• Address shortcomings in semantic and collaborative encoding
• Mitigate code assignment bias in item tokenization

Innovation

• Learnable tokenization integrating hierarchical semantics and collaborative signals
• Residual Quantized VAE with contrastive alignment for semantic regularization
• Diversity loss mitigates code assignment bias in generative recommendation