Compare two community-based personalized information recommendation algorithms

摘要

In recent years, bipartite-networks-based recommendations have attracted the attention of many researchers. Many of them are committed to improving the recommendation algorithms such as network-based inference (NBI) or probability spreading (ProbS). However, usually one or two parameters are tunable in these algorithms for optimizing the recommendation results. In these situations the optimal parameters are often applicable to specific data sets. Thus we consider using a community-based personalized recommendation, which has characteristics of simple and universal applicability. In this article, we investigate the effects of two different approaches to communities' formation based on traditional similarity formula and two improved similarity formulae proposed by us. The experimental results show that the approach of non-strictly divided communities presents greater accuracy and diversity in personalized information recommendations.