A Text Topic Discovery Method Based On Coupled Matrix And Tensor Factorization

The topic of the text reflects the main information and content of the text.Topic discovery is to discover topics of text or text set and the correlation between topics by effective methods.Topic discovery of the text is an important part of big data processing and has a wide range of applications in practice.Many effective methods have been proposed,such as clustering methods,probability based methods and so on.Since tensor can represent multidimensional data and tensor factorization can extract useful hidden information from it,tensor and tensor factorization have been widely used in data mining fields.In this paper,we create a tensor with the modes of words,time,and number of posts.In addition,a matrix is created,which is coupled with the tensor in the words mode.An optimization model of topic discovery based on coupled matrix tensor factorization is proposed,and the nonlinear conjugate gradient method is used to solve the model,and the joint factorization results of coupled matrix and tensor are obtained.Through factorization,the hidden relationship between the post data is analyzed,and the hidden topic in the post text provided by the user on the question and answer platform is obtained.Numerical results show that the topic discovery method in this paper can effectively discover the topics in the posts,and at the same time,it can also get the topic's change over time and information about the experience of users interested in a certain topic.