Improve the algorithmic performance of collaborative filtering by using the interevent time distribution of human behaviors

• The effects of interevent time of human behaviors on CF algorithm are investigated.
• The newly proposed time-related CF algorithm outperforms the standard CF algorithm.
• This work validates the findings of interest-driven model of human dynamics.

Recently, many scaling laws of interevent time distribution of human behaviors are observed and some quantitative understanding of human behaviors are also provided by researchers. In this paper, we propose a modified collaborative filtering algorithm by making use the scaling law of human behaviors for information filtering. Extensive experimental analyses demonstrate that the accuracies on MovieLensand Last.fm datasets could be improved greatly, compared with the standard collaborative filtering. Surprisingly, further statistical analyses suggest that the present algorithm could simultaneously improve the novelty and diversity of recommendations. This work provides a creditable way for highly efficient information filtering.