Improving Cold Start Recommendation by Mapping Feature-Based Preferences to Item Comparisons

Abstract

Many Recommender Systems (RSs) rely on user preference data in the form of ratings or likes for items. Previous research has shown that item comparisons can also be effectively used to model user preferences and build RS. However, users often express their preferences by referring to specific features of the items. For instance, a user may like Italian movies more than Indian ones or like action-thriller movies. In this paper, we map such preferences over features to comparisons between items. For instance, when a user's favorite feature is `action', we then assume that `action' movies are preferred to some of the movies that are not `action'. In this work we effectively incorporate these feature based comparisons in a RS and show that such preferences can be effectively combined along with other item comparisons. Moreover, we also study the usefulness of the available features.