Generalized Tag-induced Cross-Domain Collaborative Filtering

One of the most challenging problems in recommender systems based on the collaborative filtering (CF) concept is data sparseness, i.e., limited user preference data is available for making recommendations. Cross-domain collaborative filtering (CDCF) has been studied as an effective mechanism to alleviate data sparseness of one domain by transferring knowledge about user preferences from other domains. However, there are two key issues that need to be addressed to make a CDCF approach successful: (a) what common characteristics can be used to establish a link between different domains and (b) how to get each domain effectively and efficiently benefit from such a link. In this paper, we propose a novel algorithm, Generalized Tag-induced Cross-domain Collaborative Filtering (GTagCDCF), that exploits user-contributed tags as common characteristics to link different domains together. Formulated from the probabilistic point of view, GTagCDCF takes into account all the user-item relations, the user-tag relations and the item-tag relations from different domains, in which the common tags take the role of effectively transferring the knowledge between different domains. Using publicly available datasets to represent three cross-domain cases, we experimentally demonstrate that GTagCDCF substantially outperforms several state-of-the-art single domain and cross-domain CF-based recommendation approaches. GTagCDCF is also shown to be effective for heterogeneous cross-domain cases, in which different domains are characterized by different types of user preferences. In addition, our investigation of the impact of user tagging behavior on GTagCDCF led to the conclusion that users can already benefit from GTagCDCF if they only share a few common tags. Finally, we validate the robustness of GTagCDCF with respect to the scale of datasets and the number of domains, based on a three-domain experiment.