Collaborative Filtering with Attribution Alignment for Review-based Non-overlapped Cross Domain Recommendation

Cross-Domain Recommendation (CDR) has been popularly studied to utilize\ndifferent domain knowledge to solve the data sparsity and cold-start problem in\nrecommender systems. In this paper, we focus on the Review-based Non-overlapped\nRecommendation (RNCDR) problem. The problem is commonly-existed and challenging\ndue to two main aspects, i.e, there are only positive user-item ratings on the\ntarget domain and there is no overlapped user across different domains. Most\nprevious CDR approaches cannot solve the RNCDR problem well, since (1) they\ncannot effectively combine review with other information (e.g., ID or ratings)\nto obtain expressive user or item embedding, (2) they cannot reduce the domain\ndiscrepancy on users and items. To fill this gap, we propose Collaborative\nFiltering with Attribution Alignment model (CFAA), a cross-domain\nrecommendation framework for the RNCDR problem. CFAA includes two main modules,\ni.e., rating prediction module and embedding attribution alignment module. The\nformer aims to jointly mine review, one-hot ID, and multi-hot historical\nratings to generate expressive user and item embeddings. The later includes\nvertical attribution alignment and horizontal attribution alignment, tending to\nreduce the discrepancy based on multiple perspectives. Our empirical study on\nDouban and Amazon datasets demonstrates that CFAA significantly outperforms the\nstate-of-the-art models under the RNCDR setting.\n