| As one of the most common forms of data,text data covers a wide range of data,large data and uneven density.The text data stored in different platforms is different,and structured text data,semi-structured text data and unstructured text data coexist.Text mining is to extract implicit and valuable information from these massive,uneven density,and heterogeneous text data for decisions or predictions.Characteristic extraction is an indispensable key step in text mining work,the extracted features will directly affect the text mining subsequent work,the deep learning artificial neural network model for feature extraction steps can not only improve the feature extraction efficiency,can also improve the quality of complex models,in order to better reflect the text data characteristics.The main work of this article is as follows:This paper first expounds the traditional feature extraction techniques,then expounds the depth learning theory and the natural language processing technology,and analyzes the limitations of the artificial neural network model RNN,LSTM the depth learning in the text feature extraction problem.Finally,a combination of BERT pre-training model and HAN network feature extraction model is proposed.Text feature extraction is carried out from the perspective of text data preprocessing and hierarchical analysis.By setting up five groups of models,two kinds of comparative experiments are carried out on Chinese and English data sets,In order to verify the validity of the feature extraction model proposed in this paper,the experimental results show that the proposed feature extraction model has good performance in context information learning,semantic learning,word multi-layer feature learning,and has a certain weakening effect on long distance dependence. |
