| In recent years,there has been an endless stream of news about the suicide of depression patients,including some well-known public figures,which has aroused public attention to depression groups.However,the proportion of patients with depression who can receive effective treatment in our country is too low,only 5%-10%,which has become a pain point in the fight against depression.With the popularization of smartphones and the development of mobile networks,social media has become a new medium of communication.There are many user groups suffering from depression who express their opinions and record their lives in the form of text on the Internet.These textual content Contains a lot of depression-related information,which provides opportunities for active detection of depression.Therefore,this article will introduce natural language processing technology and deep learning framework,analyze the text created by Weibo users,build an automatic depression detection model,and allow hospitals,schools and other organizations to gain the initiative in the fight against depression.The specific tasks are:（1）Propose the constraints of Weibo social media user screening,and use crawler technology to crawl Chinese text corpus;（2）Propose a user-level metadata feature construction method suitable for depression detection to enhance the classification of text sequences.This feature includes three dimensions of words and grammar,readability,emotion and emotion,to explore the impact of depression on patients’ language style;（3）Exploring the sentence-level text feature extraction method based on the advanced Bert model,and propose a method for extracting user microblog text sentence vectors using the [CLS]flag;（4）Aiming at the above two types of features,a dual-input CNN model（Pseudo-siamese 2 CNN network,PS2CNN）with a pseudo-twin neural network structure is proposed to identify users with depression,and the model is applied to 500 student users in a university Tests were conducted to investigate the incidence of depression in this group,and corresponding intervention suggestions were put forward.Experiments show that the accuracy of PS2CNN’s recognition of depression users on the test set has been further improved compared to support vector machines,random forests and single convolutional neural networks,reaching 86.7%.In practical applications,the model can also effectively screen out users who are prone to depression,which is faster and more accurate than traditional diagnostic methods.The relevant departments can monitor and psychologically intervene suspected users in a timely manner,which can improve depression.The proportion of patients with effective treatment is of great significance. |
