Cross-domain Transfer of Valence Preferences via a Meta-optimization Approach

TL;DR

CVPM tackles cross-domain recommendation under data sparsity by modeling user preferences with valence-based positive and negative distributions and by learning a personalized transfer bias via a meta-learner. It integrates a common mapping with user-specific biases and employs group- and individual-level self-supervised contrastive signals to utilize non-overlapping users. The approach is validated on six transfer tasks across eight datasets, outperforming strong baselines in both value estimation and ranking, and showing robustness to overlapping-user sparsity. The work advances practical cross-domain transfer by delivering finer-grained, personalized representations and scalable learning signals, with open-source data and code.

Abstract

Cross-domain recommendation offers a potential avenue for alleviating data sparsity and cold-start problems. Embedding and mapping, as a classic cross-domain research genre, aims to identify a common mapping function to perform representation transformation between two domains. Nevertheless, previous coarse-grained preference representations, non-personalized mapping functions, and excessive reliance on overlapping users limit their performance, especially in scenarios where overlapping users are sparse. To address aforementioned challenges, we propose a novel cross-domain approach, namely CVPM. CVPM formalizes cross-domain interest transfer as a hybrid architecture of parametric meta-learning and self-supervised learning, which not only transfers user preferences at a finer level, but also enables signal enhancement with the knowledge of non-overlapping users. Specifically, with deep insights into user preferences and valence preference theory, we believe that there exists significant difference between users' positive preferences and negative behaviors, and thus employ differentiated encoders to learn their distributions. In particular, we further utilize the pre-trained model and item popularity to sample pseudo-interaction items to ensure the integrity of both distributions. To guarantee the personalization of preference transfer, we treat each user's mapping as two parts, the common transformation and the personalized bias, where the network used to generate the personalized bias is output by a meta-learner. Furthermore, in addition to the supervised loss for overlapping users, we design contrastive tasks for non-overlapping users from both group and individual-levels to avoid model skew and enhance the semantics of representations. Exhaustive data analysis and extensive experimental results demonstrate the effectiveness and advancement of our proposed framework.

Figures (11)

• Figure 1: Differences between CVPM and previous work. Previous work utilize a common mapping function, while ours is individualized.
• Figure 2: The overall framework of CVPM. (a) Pretraining using MF. (b) Sampling enhancment. (c) Distribution learning. (d) Generic bridge and personalized bias network. Blue, orange, and green represent users, positive items, and negative items, respectively.
• Figure 3: Popularity-guild dynamic sampling. We adopt different heuristic sampling strategies for positive and negative candidate sets.
• Figure 4: Model optimization. (a) Supervision signals for overlapping users. (b) Group-level contrastive loss. (c) Individual-level contrastive loss. Stars of different colors represent the cluster centroids of the source domain.
• Figure 5: Data analysis.