Research on HMM based link prediction method in heterogeneous network

Abstract

In order to solve the problem that incomplete mining of structural information and semantic information in heterogeneous networks, a link prediction method combining meta-path-based analysis and hidden Markov model was proposed for link prediction of heterogeneous network.Considering that clustering could effectively capture the structural information of heterogeneous network, the k-means algorithm was improved to obtain the initial clustering center method based on the minimum distance mean square error, and it was applied to the hidden Markov model, first-order cluster hidden markov model (C-HMM(1)) link prediction method, and a link prediction method for heterogeneous network with second-order cluster hidden Markov model (C-HMM(2)) were designed.Further, considering the feature information of the data, a link prediction method called ME-HMM that combined the maximum entropy model and the second-order Markov model was proposed.The experimental results show that the ME-HMM has higher link prediction accuracy than the C-HMM, and the ME-HMM method has better performance than the C-HMM method because it fully considers the feature information of the data.

Keywords

• heterogeneous network;link prediction;hidden Markov model;clustering;maximum entropy