Collaborative Context-aware Preference Learning

Preference learning methods work by exploiting patterns in the data that relate users to items. Preference data often includes information such as the context of a recommendation (e.g. time/date, location). Leveraging this data (e.g. click logs, purchase/usage data) can significantly improve the relevance and quality of the recommendation. In this work we introduce a novel scalable context-aware collaborative filtering approach that is based on Tensor Factorization where additional information is represented by additional dimensions in the Tensor. The algorithm is tested on data from the Android mobile application recommendation service appazaar1 and from the Last.fm2 music service where it compares favorably with state-of-the-art collaborative filtering methods.