Subprofile-Aware Diversification of Recommendations
Refereed Original Article
A user of a recommender system is more likely to be satisfied by one or more ofthe recommendations if each individual recommendation isrelevantto her but addi-tionally if the set of recommendations isdiverse. The most common approach torecommendation diversification uses re-ranking: the recommender system scores aset of candidate items for relevance to the user; it then re-ranks the candidates so thatthe subset that it will recommend achieves a balance between relevance and diversity.Ordinarily, we expect a trade-off between relevance and diversity: the diversity of theset of recommendations increases by including items that have lower relevance scoresbut which are different from the items already in the set. In early work, the diversityof a set of recommendations was given by the average of their distances from oneanother, according to some semantic distance metric defined on item features suchas movie genres. More recentintent-awarediversification methods formulate diver-sity in terms of coverage and relevance ofaspects. The aspects are most commonlydefined in terms of item features. By trying to ensure that the aspects of a set of recom-mended items cover the aspects of the items in the user’s profile, the level of diversityis more personalized. In offline experiments on pre-collected datasets, intent-awarediversification using item features as aspects sometimes defies the relevance/diversitytrade-off: there are configurations in which the recommendations exhibits increasesin both relevance and diversity. In this paper, we present a new form of intent-awarediversification, which we call SPAD (Subprofile-Aware Diversification), and a variantcalled RSPAD (Relevance-based SPAD). In SPAD, the aspects are not item features;they aresubprofilesof the user’s profile. We present and compare a number of differ-ent ways to extract subprofiles from a user’s profile. None of them is defined in termsof item features. Therefore, SPAD is useful even in domains where item features arenot available or are of low quality. On three pre-collected datasets from three dif-ferent domains (movies, music artists and books), we compare SPAD and RSPAD tointent-aware methods in which aspects are item features. We find on these datasets thatSPAD and RSPAD suffer even less from the relevance/diversity trade-off: across allthree datasets, they increase both relevance and diversity for even more configurationsthan other approaches to diversification. Moreover, we find that SPAD and RSPADare the most accurate systems across all three datasets.
Digital Object Identifer (DOI):
User Modeling and User-Adapted Interaction
National University of Ireland, Cork (UCC)
Open access repository: