{"ID":2850526,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2510.21242","arxiv_id":"2510.21242","title":"Bi-Level Optimization for Generative Recommendation: Bridging Tokenization and Generation","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 multiple 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. We release our code at https://github.com/Ten-Mao/BLOGER.","short_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 id...","url_abs":"https://arxiv.org/abs/2510.21242","url_pdf":"https://arxiv.org/pdf/2510.21242v2","authors":"[\"Yimeng Bai\",\"Chang Liu\",\"Yang Zhang\",\"Dingxian Wang\",\"Frank Yang\",\"Andrew Rabinovich\",\"Wenge Rong\",\"Fuli Feng\"]","published":"2025-10-24T08:25:56Z","proceeding":"cs.IR","tasks":"[\"cs.IR\"]","methods":"[]","has_code":false,"code_links":[{"ID":607808,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2850526,"paper_url":"https://arxiv.org/abs/2510.21242","paper_title":"Bi-Level Optimization for Generative Recommendation: Bridging Tokenization and Generation","repo_url":"https://github.com/Ten-Mao/BLOGER","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}