Chinese Depression Text Classification Based on Domain Emotion Dictionary and Word Feature Fusion

Abstract

Aiming at the accurate classification of Chinese comment corpus on depressive tendencies, a text classification method of Chinese depression based on the fusion of domain emotion dictionary and word features was proposed. Firstly, based on the general domain dictionary and word similarity calculation, the Chinese emotion dictionary in the field of depression was constructed and extended; Secondly, according to the different characteristics of Chinese words, the representation ability of words was optimized by adding emotional phrases to the word level model and combining the word characteristic and phrase characteristic; Finally, the method was applied to the BERT model to construct a BERT-W model for Chinese depression text classification task. The experimental results show that compared with the representation model based on shallow neural network, the BERT-W model performs better in the accuracy, recall and F1 values of CDSD data sets; Compared with the BERT-char model and the BERT-word model based on words or whole domain words, the BERT-W model integrating word level features and emotional word features has stronger representation ability, with an accuracy rate of 97. 59% and an F1 value of 96.11%. It has a higher processing ability in the classification of complex corpora of patients with depression, and is conducive to the subsequent classification of complex corpora of patients with depression.