A Content-Based Recommendation System Using TrueSkill

Abstract

We present a probabilistic approach based on TrueSkill for Content-Based Recommendation Systems. On one hand, this proposal allow us to tackle the "cold start" problem because it relies on a content-based approach. On the other hand, it is valuable for handling high uncertainty since it solely depends on available items and ratings given by users. Thus, there is no dependency on the number of items and users. In addition, it is highly scalable because user preferences get richer as items get ranked.