Keyword extraction method of text combining random walk and variational inference

Keyword extraction is an important way to explore text semantic information. It is mainly composed of topic model and parameter estimation algorithm. Firstly, the mapping relationship between topic space and text space is constructed by topic model, and then the model parameters are estimated by expectation maximization (EM) algorithm. The traditional EM algorithm constructs the logarithmic likelihood function of topic variables as the target expectation, and maximizes it by Lagrange multiplier method. In terms of expectation construction, this method does not consider setting an approximate distribution for the posterior distribution of topic variables, and in the aspect of programming solution, a large number of partial derivative operations are required, which makes it difficult to analyze the convergence of the algorithm. In this paper, a keyword extraction method combining random walk and variational inference is proposed. The variational parameter distribution is set to estimate the posterior distribution of topic variables, and the random walk state transition matrix is used to iteratively calculate the parameters instead of Lagrange function. Not only can it be proved theoretically that this method has a stable distribution, but also through comparative experiments, it is proved that this method is superior to the ordinary EM algorithm in three evaluation metrics:precision, recall and F1 Measure, and its performance is not easily affected by the change of the number of extracted words.