Research On The Classification Method Of Social Governance Text

The social governance text is an important basis for the government to carry out social management.It contains the public’s reflection of various problems in society and contains rich social values.Therefore,it is very important to accurately extract the hot issues that people care about.However,these text data are relatively large in size and very cumbersome to process.In addition,these texts have the characteristics of short length and variable content length,and may even contain many redundant texts without topic representation capabilities.This makes it difficult to understand the current public opinion hotspots.Text classification methods based on text mining technology are designed to find social hotspot issues in this paper.The LDA topic model is a semantic mining method,which is used here to classify social governance texts,infer the parameters and semantic information of the mixed distribution of topics in the texts,so as to identify and classify information such as topics of new texts.After analyzing and evaluating the experimental results,a classification method combining LDA and bayesian networks is proposed.LDA is first used to expand the features of social governance texts,and then the bayesian network is used to classify the expanded text,so that the text classification effect is improved.In the research of text classification,algorithms in the field of deep learning have been widely used by experts and scholars,and have achieved better performance than traditional machine learning algorithms in some classification tasks.Deep learning can actively learn the syntactic and semantic features of the text,and then obtain the deep features of the text information,reduce the difficulty of manual feature extraction,and have objectivity.Therefore,the Bert model based on deep learning is used for classification research on social governance texts in this paper.However,due to the characteristics of the text data such as short text content,irregular syntax structure and text category crossing,the text classification effect of Bert is not ideal.Therefore,a text classification method combining Bert model and bayesian network is put forward.The bayesian network is first used in a binary classification experiment to distinguish the livelihood query class from other categories,and then the Bert model is used to specifically classify the text under other categories.The effect of text classification has been improved.