Personalized Recommendation via Enhanced Redundant Eliminated Network-Based Inference

An efficient recommendation system is fundamental to solve the problem of information overload in modern society. In physical dynamics, material diffusion based on binary networks has a wide range of applications in recommendation systems. However, material diffusion has the problem of excessive diffusion and redundant similarity. In the past studies, most people focused on reducing the popularity of popular items. This paper mitigates the problem of redundant similarity by considering the second-order similarity of the enhanced items. It evaluates the algorithm through three real datasets (MovieLens, Netffix and RYM), which proves the method is superior to other algorithms in accuracy, diversity and novelty.