An adaptive approach to dealing with unstable behaviour of users in collaborative filtering systems

Recommendation systems manage information overload in order to present personalized content to users based on their interests. One of the most efficient recommendation approaches is collaborative filtering, through which recommendation is based on previously rated data. Collaborative filtering techniques feature impressive solutions for suggesting favourite items to certain users. However, recommendation methods fail to reflect fluctuations in users' behaviour over time. In this article, we propose an adaptive collaborative filtering algorithm which takes time into account when predicting users' behaviour. The transitive relationship from one user to another is considered when computing the similarity of two different users. We predict variations of users' preferences using their profiles. Our experimental results show that the proposed algorithm is more accurate than the classical collaborative filtering technique.