Accurate and diverse recommendations using item-based subprofiles

In many approaches to recommendation diversification, a recommender scores items for relevance and then re-ranks them to balance relevance with diversity. In intent-aware diversification, diversity is formulated in terms of coverage of aspects, where aspects are either explicit such as movie genres or implicit such as the latent factors found during matrix factorization. Typically, the same set of aspects is used across all users. In this paper, we propose a form of intent-aware diversification, which we call SPAD (SubProfile-Aware Diversification), and a variant called RSPAD (Relevance-based SPAD). The aspects we use in SPAD and RSPAD are subpro-files of the user's profile. They are not defined in terms of explicit or implicit features. We compare our methods to other forms of intent-aware diversification. We find that SPAD and RSPAD always improve accuracy (as measured by precision) and diversity (as measured by a-nDCG) even though the diversity metric in our experiments uses explicit features but SPAD and RSPAD make no use of them.