MCRPL: A Pretrain, Prompt & Fine-tune Paradigm for Non-overlapping Many-to-one Cross-domain Recommendation

TL;DR

This work tackles non overlapping many-to-one cross domain recommendation by proposing MCRPL, a prompt enhanced two stage learning framework that learns a shared domain knowledge through domain agnostic prompts and captures domain specific nuances via domain specific prompts. The model pre trains on all available domains to embed common knowledge, then fine tunes on the sparse target domain by freezing the shared components and adapting only the domain specific prompts, aided by an orthogonal loss to separate shared and target specific information. Empirical results on two real world datasets show significant improvements over single domain and cross domain baselines, with ablations confirming the critical role of pre training, prompt design, and the two stage training paradigm. The approach reduces negative transfer in non overlapping settings and demonstrates strong scalability and applicability to sequential recommendation tasks, with potential extensions to rating and CTR prediction.

Abstract

Cross-domain Recommendation (CR) is the task that tends to improve the recommendations in the sparse target domain by leveraging the information from other rich domains. Existing methods of cross-domain recommendation mainly focus on overlapping scenarios by assuming users are totally or partially overlapped, which are taken as bridges to connect different domains. However, this assumption does not always hold since it is illegal to leak users' identity information to other domains. Conducting Non-overlapping MCR (NMCR) is challenging since 1) The absence of overlapping information prevents us from directly aligning different domains, and this situation may get worse in the MCR scenario. 2) The distribution between source and target domains makes it difficult for us to learn common information across domains. To overcome the above challenges, we focus on NMCR, and devise MCRPL as our solution. To address Challenge 1, we first learn shared domain-agnostic and domain-dependent prompts, and pre-train them in the pre-training stage. To address Challenge 2, we further update the domain-dependent prompts with other parameters kept fixed to transfer the domain knowledge to the target domain. We conduct experiments on five real-world domains, and the results show the advance of our MCRPL method compared with several recent SOTA baselines.

Figures (6)

• Figure 1: An example that illustrates the drawbacks of traditional cross-domain methods in the non-overlapping scenarios.
• Figure 2: The overall system architecture of MCRPL. For each item in the inputs, we enhance its representation by the prompt layer.
• Figure 3: Impact of different fine-tuning strategies on both datasets.
• Figure 4: Impact of the hyper-parameter $\lambda$ on HVIDEO and MIXED datasets.
• Figure 5: Impact of the prompt length $L_{p}$ on HVIDEO and MIXED datasets.