Bi-Level Optimization for Generative Recommendation: Bridging Tokenization and Generation

2025/10/24 by Bai, Yimeng, Liu, Chang, Zhang, Yang +5 · 4 citations
#FOS: Computer and information sciences #H.3.3 #H.3.5 #Information Retrieval (cs.IR)

paper · doi:10.48550/arxiv.2510.21242

Abstract

Generative recommendation is emerging as a transformative paradigm by directly generating recommended items, rather than relying on matching. Building such a system typically involves two key components: (1) optimizing the tokenizer to derive suitable item identifiers, and (2) training the recommender based on those identifiers. Existing approaches often treat these components separately--either sequentially or in alternation--overlooking their interdependence. This separation can lead to misalignment: the tokenizer is trained without direct guidance from the recommendation objective, potentially yielding suboptimal identifiers that degrade recommendation performance. To address this, we propose BLOGER, a Bi-Level Optimization for GEnerative Recommendation framework, which explicitly models the interdependence between the tokenizer and the recommender in a unified optimization process. The lower level trains the recommender using tokenized sequences, while the upper level optimizes the tokenizer based on both the tokenization loss and recommendation loss. We adopt a meta-learning approach to solve this bi-level optimization efficiently, and introduce gradient surgery to mitigate gradient conflicts in the upper-level updates, thereby ensuring that item identifiers are both informative and recommendation-aligned. Extensive experiments on real-world datasets demonstrate that BLOGER consistently outperforms state-of-the-art generative recommendation methods while maintaining practical efficiency with no significant additional computational overhead, effectively bridging the gap between item tokenization and autoregressive generation.

Citations

• OneRec-Think: In-Text Reasoning for Generative Recommendation
• IGD: Token Decisiveness Modeling via Information Gain in LLMs for Personalized Recommendation
• Reinforced Latent Reasoning for LLM-based Recommendation
• OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment
• Unified Semantic and ID Representation Learning for Deep Recommenders
• Generative Recommender with End-to-End Learnable Item Tokenization
• EAGER: Two-Stream Generative Recommender with Behavior-Semantic Collaboration
• TokenRec: Learning to Tokenize ID for LLM-based Generative Recommendation
• Text-like Encoding of Collaborative Information in Large Language Models for Recommendation
• Learnable Item Tokenization for Generative Recommendation
• IDGenRec: LLM-RecSys Alignment with Textual ID Learning
• Adapting Large Language Models by Integrating Collaborative Semantics for Recommendation
• CoLLM: Integrating Collaborative Embeddings into Large Language Models for Recommendation
• CoLLM: Integrating Collaborative Embeddings Into Large Language Models for Recommendation
• Generative Retrieval with Semantic Tree-Structured Item Identifiers via Contrastive Learning
• A Bi-Step Grounding Paradigm for Large Language Models in Recommendation Systems
• Llama 2: Open Foundation and Fine-Tuned Chat Models
• Recommender Systems with Generative Retrieval
• Generative Recommendation: Towards Next-generation Recommender Paradigm
• Efficient Bi-Level Optimization for Recommendation Denoising
• Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)
• Feedback Loop and Bias Amplification in Recommender Systems
• LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation
• LightGCN
• Gradient Surgery for Multi-Task Learning
• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• Meta-Learning with Implicit Gradients
• Hierarchical Gating Networks for Sequential Recommendation
• Self-Attentive Sequential Recommendation
• Meta-learning with differentiable closed-form solvers
• Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Matrix Factorization Techniques for Recommender Systems

Cited by

• Reward Guided Decoding for Generative Recommendation
• Grevo: A Unified Generative Recommendation Framework with Evolutionary Item Indexing
• Multi-Decoder OneRec: Controllable Generative Retrieval for Multi-Objective Industrial Recommendation
• Collaborative Memory Augmentation for Generative Recommendation

Related

• Generative Recommender with End-to-End Learnable Item Tokenization
• Pre-training Generative Recommender with Multi-Identifier Item Tokenization
• Learnable Item Tokenization for Generative Recommendation
• CoFiRec: Coarse-to-Fine Tokenization for Generative Recommendation
• Closing the Performance Gap in Generative Recommenders with Collaborative Tokenization and Efficient Modeling