This paper addresses the problem of multi-objective cross-domain recommendation (MTCDR), which aims to simultaneously improve recommendation performance across multiple domains. Existing MTCDR methods primarily rely on domain-shared entities (e.g., users or items) to fuse and transfer cross-domain knowledge, but they are ineffective in non-overlapping recommendation scenarios. Some studies address MTCDR by modeling user preferences and item features as domain-shared semantic representations, but this requires extensive auxiliary data for pretraining. Inspired by recent advances in generative recommendation, this paper presents GMC, a generative paradigm-based MTCDR approach. GMC integrates multi-domain knowledge within a unified generative model using semantically quantized discrete item identifiers. An item tokenizer is used to generate domain-shared semantic identifiers for each item, and a domain-integrated sequence-to-sequence model is trained to formulate item recommendation as a token generation task. To improve performance, domain-aware contrastive loss is incorporated into semantic identifier learning, and domain-specific fine-tuning of the unified recommender system is performed. Extensive experiments on five public datasets demonstrate the effectiveness of GMC compared to several baseline methods.
