Active learning in multi-domain collaborative filtering recommender systems

The lack of information is an acute challenge in most recommender systems, especially for the collaborative filtering algorithms which utilize user-item rating matrix as the only source of information. Active learning can be used to remedy this problem by querying users to give ratings to some items. Apart from the active learning algorithms, cross-domain recommender system techniques try to alleviate the sparsity problem by exploiting knowledge from auxiliary (source) domains. A special case of cross-domain recommendation is multi-domain recommendation that utilizes the shared knowledge across multiple domains to alleviate the data sparsity in all domains. In this paper, we propose a novel multi-domain active learning framework by incorporating active learning techniques with cross-domain collaborative filtering algorithms in the multi-domain scenarios. Specifically, our proposed active learning elicits all the ratings simultaneously based on the criteria with regard to both items and users, for the purpose of improving the performance of the whole system. We evaluate a variety of active learning strategies in the proposed framework on different multi-domain recommendation tasks based on three popular datasets: Movielens, Netflix and Book-Crossing. The results show that the system performance can be improved further when combining cross-domain collaborative filtering with active learning algorithms.