Contrastive Learning for Extracting Transferable User Profiles in Cross-Domain Recommendation System
Qingquan Zhao, Qi Wang · 2024
The cold start issue represents a major obstacle in recommendation systems. Nevertheless, engaging new information within a supplementary domain can help generate accurate recommendations for them in the specific domain of interest. The essence of Cross Domain Recommendation (CDR) is the transfer of user preferences from one domain to another, offering a potential solution to the cold start problem. The majority of contemporary techniques primarily utilize a bridging mechanism to facilitate the transfer of information between different domains. However, due to the uniqueness of each user’s personality, their corresponding characteristics also vary. The combination of individual user traits with domain information further complicates the cross-domain transfer process. This complexity results in challenges in achieving precise and effective information transfer across domains in a single mapping step. Concurrently, it is noteworthy that the majority of existing methods predominantly center on the similarity relationships between two domains, neglecting to extract the distinct features of users through contrastive analysis. In our study, we introduce an innovative approach for Cross-Domain Recommendation Systems, termed as the Contrastive Learning for Extracting Transferable User Profiles (CLUPCDR) framework. This framework primarily employs contrastive learning techniques to develop a comparative strategy, which is adept at producing user features with higher transferability. These features are particularly optimized for cross-domain knowledge application. Additionally, we incorporate a unique validation methodology to ascertain the improved cross-domain applicability of these user features. To demonstrate the e cacy of our CLUPCDR framework, comprehensive experiments were conducted on two diverse real-world datasets, each consisting of three distinct sub-datasets. The code has been available at https://github.com/AnonymousAuthorAloha/CLUPCDR.git.