Extracting the Information Backbone Based On Personalized Time Window

Research on recommendation systems in bipartite networks has mainly been dedicated to enhance the accuracy of recommendations while neglecting the fact that complete historical information can be redundant or even mislead to the recommendations. In this paper, we first investigate the impact of the time window on recommendation models. We gradually expand the time window and find that the performance remains almost unchanged. We set the size of the time window according to the user’s temporal and topological information; thus, different users have information backbones of different sizes. The experimental results on real networks show that the computational time complexity can be improved by the algorithm while simultaneously decreasing the data storage requirements.