Hybrid Text Topic Discovery Method for Multi-source Information

Abstract

With the wide application of information technology and the rapid development of cyberspace, cyberspace security has become a hot topic in society. Topic discovery technology plays a vital role to grasp the direction of public opinion in the governance of cyberspace security. Due to the diversity of information sources, the distribution of hybrid text presents inhomogeneous characteristics. Most of the current topic discovery problems are addressed by clustering algorithms, which are not effective in processing hybrid text. Therefore, the paper proposes a multi-source text clustering method based on the Dirichlet Multinomial Mixture model (DMM). By considering the characteristic difference between multi-source text data, we propose a feature fusion algorithm based on the TextRank algorithm. Furthermore, the problem of text sparse and high-dimensionality in feature fusion is handled by the DMM model. Numerical results reveal that the proposed clustering algorithm significantly improves the multi-source text clustering performance, which relieves the influence of text feature inhomogeneous, text sparse and high-dimensionality after feature fusion. It provides decision support for the governance of cyberspace security and contributes to the stable development of society.