Highly effective recommender systems may still face users’ interest drifting. One of the main strategies for handling interest-drifting is forgetting mechanism. Current approaches based on forgetting mechanism have some drawbacks: (i) Drifting times are not considered to be detected in user interest over time. (ii) They are not adaptive to the evolving nature of user’s interest. Until now, there hasn’t been any study to overcome these problems. This paper discusses the above drawbacks and presents a novel recommender system, named WmIDForg, using web usage mining, web content mining techniques, and forgetting mechanism to address user interest-drift problem. We try to detect evolving and time-variant patterns of users' interest individually, and then dynamically use this information to predict favorite items of the user better over time. The experimental results on EachMovie dataset demonstrate our methodology increases recommendations precision 6.80% and 1.42% in comparison with available approaches with and without interest-drifting respectively.
Rights and permissions | |
![]() | This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |