Exploring Preference-Guided Diffusion Model for Cross-Domain Recommendation

TL;DR

The paper tackles the cold-start challenge in cross-domain recommendation by shifting from a purely embedding-and-mapping paradigm to a diffusion-based approach that explicitly injects source-domain user preferences into target-domain representations. DMCDR uses a Transformer-based preference encoder to produce a guidance signal and a diffusion process that progressively denoises a user representation in the target domain, guided at each step by the source-domain signal, with classifier-free guidance to improve alignment. Empirical results across three real-world CDR scenarios show DMCDR delivering substantial improvements over state-of-the-art baselines (up to around 32.6%), validating the effectiveness of explicit preference injection via diffusion and highlighting the importance of components like the preference encoder, guidance signal, and diffusion module. This work introduces a new direction for cross-domain recommendation by leveraging diffusion models to enable robust transfer of user preferences, especially in cold-start settings, and suggests avenues for extending to multi-domain settings and more complex user-item contexts.

Abstract

Cross-domain recommendation (CDR) has been proven as a promising way to alleviate the cold-start issue, in which the most critical problem is how to draw an informative user representation in the target domain via the transfer of user preference existing in the source domain. Prior efforts mostly follow the embedding-and-mapping paradigm, which first integrate the preference into user representation in the source domain, and then perform a mapping function on this representation to the target domain. However, they focus on mapping features across domains, neglecting to explicitly model the preference integration process, which may lead to learning coarse user representation. Diffusion models (DMs), which contribute to more accurate user/item representations due to their explicit information injection capability, have achieved promising performance in recommendation systems. Nevertheless, these DMs-based methods cannot directly account for valuable user preference in other domains, leading to challenges in adapting to the transfer of preference for cold-start users. Consequently, the feasibility of DMs for CDR remains underexplored. To this end, we explore to utilize the explicit information injection capability of DMs for user preference integration and propose a Preference-Guided Diffusion Model for CDR to cold-start users, termed as DMCDR. Specifically, we leverage a preference encoder to establish the preference guidance signal with the user's interaction history in the source domain. Then, we explicitly inject the preference guidance signal into the user representation step by step to guide the reverse process, and ultimately generate the personalized user representation in the target domain, thus achieving the transfer of user preference across domains. Furthermore, we comprehensively explore the impact of six DMs-based variants on CDR.

Figures (7)

• Figure 1: An illustration of (a) traditional CDR methods, where $\otimes$ denotes the integration operation; (b) our DMCDR.
• Figure 2: The overall framework of DMCDR in the training phase. To achieve preference transfer, DMCDR encodes user's interaction history $\mathcal{H}_i^s$ in the source domain as preference guidance signal $\bm{h}_i^s$, then explicitly injects $\bm{h}_i^s$ into the user representation step by step to guide the reverse process, and finally generates the personalized user representation in the target domain.
• Figure 3: An illustration of six DMs-based variants in the training phase, where $\oplus$ denotes the concatenation operation.
• Figure 4: Effect of inference step $T'$ and guidance strength $\omega$.
• Figure 5: Effect of noise scale $\eta$ and diffusion step $T$.