Proposal of Personal Value-Based User Modeling Using Latent Factors

This paper proposes a method for modeling users’ personal values as latent factors through matrix factorization. The proposed method is applied to memory-based collaborative filtering (CF) and Factorization Machines (FM) to demonstrate its effectiveness. A recent trend in CF is to introduce additional factors than interaction history. A rate matching rate (RMRate) has been proposed for modeling user’s personal values, and it has been shown to be effective for increasing diversity and recommending niche (long-tail or unpopular) items. On the other hand, the disadvantage of RMRate is that it needs an attribute-level evaluations in addition to rating (total evaluation) to items, which limits its applicability. To address this problem, personal value-based modeling method without using attribute-level evaluations has been proposed, which defines user’s personal values as their tendency to select popular/unpopular items. Experimental results showed that this method was effective especially for users who rated a small number of items. However, this approach is difficult to extend for obtaining other types of personal values. To obtain different types of personal values of users only from a rating matrix, this paper proposes to calculate RMRate by regarding the product of user’s and item’s latent vectors as a pseudo attribute-level evaluation. Experimental results with two datasets show the RMRate calculated by the proposed method can improve precision, recall, and nDCG of memory-based CF and FM.