An inverse collaborative filtering approach for cold-start problem in web service recommendation

Abstract

Due to increasing volume and variety of web in different service communities, users are apt to find their interested web through various recommendation techniques, e.g., Collaborative Filtering (i.e., CF) recommendation. In CF (e.g., user-based CF, item-based CF or hybrid CF) recommendation, the similar of target or the similar of target (i.e., preferred by target user) are determined first; afterwards, the preferred by similar friends or the similar of target services are recommended to target user. However, due to inherent data sparsity in service recommendation, cold-start problem is inevitable when target user has no similar and target have no similar services. While present CF recommendation approaches cannot deal with this cold-start problem very well. In view of this shortcoming, in this paper, a novel inverse CF approach named Inverse_CF_Rec is introduced to help alleviate cold-start problem in service recommendation. Concretely, in Inverse_CF_Rec, we first look for target user's enemy (i.e., antonym of friend), and then determine target user's possible friends based on Social Balance Theory (e.g., enemy's enemy is a friend rule). Afterwards, the preferred by possible friends of target or the disliked by enemies of target are recommended to target user, so as to alleviate cold-start problem. Finally, through a set of simulation experiments deployed on well-known MovieLens-1M dataset, we validate feasibility of our proposal in terms of recommendation accuracy, recall and efficiency.