Topic Analysis of Chinese Documents Based on Key Phrases and Latent Dirichlet Allocation Model

In order to overcome the problems of limited semantic understanding, inaccurate topic calibration, and ignoring the correlation between topics of keywords in document topic analysis, this paper proposes a document topic analysis model based on key phrases. In the process of generating a candidate set of key phrases, an improved SegPhrase algorithm is used to add mutual information features between word strings. In the process of evaluating the quality of phrases, different features are given different weights to comprehensively evaluate the phrases, which can retain some low-frequency but key phrases. Then, the key phrases are combined with LDA topic analysis model to extract topic information from the text. Through experiments, it has been proven that key phrases under each topic have a higher correlation with the topic, and can express document topic information more clearly and accurately. However, the correlation between individual words and the topic is relatively low.