Abstract
Recently, there has been a surge of interest in Multi-Target Cross-DomainRecommendation (MTCDR), which aims to enhance recommendation performance acrossmultiple domains simultaneously. Existing MTCDR methods primarily rely ondomain-shared entities (\eg users or items) to fuse and transfer cross-domainknowledge, which may be unavailable in non-overlapped recommendation scenarios.Some studies model user preferences and item features as domain-sharablesemantic representations, which can be utilized to tackle the MTCDR task.Nevertheless, they often require extensive auxiliary data for pre-training.Developing more effective solutions for MTCDR remains an important area forfurther exploration. Inspired by recent advancements in generative recommendation, this paperintroduces GMC, a generative paradigm-based approach for multi-targetcross-domain recommendation. The core idea of GMC is to leverage semanticallyquantized discrete item identifiers as a medium for integrating multi-domainknowledge within a unified generative model. GMC first employs an itemtokenizer to generate domain-shared semantic identifiers for each item, andthen formulates item recommendation as a next-token generation task by traininga domain-unified sequence-to-sequence model. To further leverage the domaininformation to enhance performance, we incorporate a domain-aware contrastiveloss into the semantic identifier learning, and perform domain-specificfine-tuning on the unified recommender. Extensive experiments on five publicdatasets demonstrate the effectiveness of GMC compared to a range of baselinemethods.