Collaborative filtering inspired from language modeling

Abstract

Recommender systems filter resources for a given user by predicting the most pertinent item given a specific context. This paper describes a new approach of generating suitable recommendations based on the active user's navigation stream. The underlying hypothesis is that the items order in the stream results from the intrinsic logic of the user's behavior. We show similarities between natural language and Internet navigation and put forward navigation specificities. We then design a new model that integrates advantages of statistical language models such as n-grams and triggers to compute recommendations. The resulting Sequence Based Recommender has been tested on Internet navigation artificial corpora.