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

2025-10-24

AI Summary

In generative recommendation, item tokenization and recommendation modeling are typically decoupled, leading to tokenizers lacking recommendation-aware guidance and causing misalignment between token informativeness and recommendation consistency. To address this, we propose BLOGER—a novel framework that jointly optimizes tokenization and generative recommendation as a bi-level optimization problem: the upper level adapts the tokenizer to produce recommendation-friendly tokens, while the lower level trains the generative recommender. To mitigate gradient conflicts between levels, we introduce a gradient surgery mechanism and employ meta-learning for efficient optimization. BLOGER enables end-to-end joint training within an autoregressive generation paradigm. Extensive experiments on multiple real-world datasets demonstrate that BLOGER significantly outperforms state-of-the-art methods in recommendation accuracy, with consistent performance gains and manageable computational overhead. It effectively bridges the semantic gap between tokenization and generative recommendation.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• 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

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.

Problem

• Optimizing tokenizer and recommender interdependence in generative recommendation systems
• Addressing misalignment between tokenization quality and recommendation performance
• Developing unified bi-level optimization for joint tokenizer-recommender training

Innovation

• Bi-level optimization unifies tokenizer and recommender training
• Meta-learning approach efficiently solves bi-level optimization problem
• Gradient surgery mitigates conflicts between tokenization and recommendation objectives