DisCo: Graph-Based Disentangled Contrastive Learning for Cold-Start Cross-Domain Recommendation

TL;DR

DisCo tackles negative transfer in cold-start cross-domain recommendation by learning disentangled user intents with a multi-channel graph encoder and high-order user similarity via affinity graphs. It then performs intent-wise intra- and inter-domain contrastive learning to bridge domains while preserving target-domain signals, guided by a cross-domain decoder. Experiments on four Amazon domain pairs show DisCo consistently outperforms state-of-the-art baselines, validating the effectiveness of its disentangled and contrastive components. The approach provides a principled way to filter irrelevant source-domain information and improve cold-start recommendations across domains.

Abstract

Recommender systems are widely used in various real-world applications, but they often encounter the persistent challenge of the user cold-start problem. Cross-domain recommendation (CDR), which leverages user interactions from one domain to improve prediction performance in another, has emerged as a promising solution. However, users with similar preferences in the source domain may exhibit different interests in the target domain. Therefore, directly transferring embeddings may introduce irrelevant source-domain collaborative information. In this paper, we propose a novel graph-based disentangled contrastive learning framework to capture fine-grained user intent and filter out irrelevant collaborative information, thereby avoiding negative transfer. Specifically, for each domain, we use a multi-channel graph encoder to capture diverse user intents. We then construct the affinity graph in the embedding space and perform multi-step random walks to capture high-order user similarity relationships. Treating one domain as the target, we propose a disentangled intent-wise contrastive learning approach, guided by user similarity, to refine the bridging of user intents across domains. Extensive experiments on four benchmark CDR datasets demonstrate that DisCo consistently outperforms existing state-of-the-art baselines, thereby validating the effectiveness of both DisCo and its components.

Figures (5)

• Figure 1: An illustration of user similarity distortion in CDR. Users may share diverse intents across domains. As a result, for the target user $u$, some users (i.e., $u_1$, $u_3$) may have similar preferences in the source domain but differ significantly in the target domain, leading to negative transfer.
• Figure 2: Illustration of the proposed framework DisCo.
• Figure 3: Performance comparison w.r.t. different numbers of user intent $K$.
• Figure 4: Performance comparison w.r.t. different values of $d$ and $\alpha$ for random walks and high-order user similarity.
• Figure 5: Performance comparison w.r.t. different values of $\lambda$ and $\beta$ for the overall objective.