Research On Depression Prediction Based On Capsule Network Model

Depression has become the fourth largest disease in the world.Due to the difficulty in diagnosis of mental problems,the current diagnosis method is mainly based on questionnaire survey,supplemented by the judgment of doctors.Its accuracy depends not only on the doctor’s expertise and experience,but also on the patient’s cooperation.In addition,early diagnosis and evaluation of patients are limited.Because patients are not aware of their disease,and are not willing to seek medical treatment.As a result,treatment rates for depression are low.With the development of economy and technology,social media is used by more and more people.They tend to record their daily routines and moods on social media,which creates a wealth of data for researchers to use.However,depression prediction is not the same as text classification.First of all,there is no universal standard for determining depression.In addition,deep learning models for identifying depression are not easy to be built.To solve these problems,a capsule network model combining local and global features is proposed.This model integrates depression dictionary and capsule network,which can effectively find the potential depression patients in Weibo users.The main work is as follows:(1)Acquisition and preprocessing of data sets: Firstly,the data of depressed users and non-depressed users on microblogs are obtained.After that,the original data needs to be preprocessed.Specific operations include selecting appropriate micro blog texts and eliminating redundant information such as punctuation marks,emoticons and numbers.Finally,the word segmentation tool is used for word segmentation,and each word is mapped into a multi-dimensional word vector.(2)Construction of a depression dictionary for extracting local features: Through the analysis of the existing emotion dictionaries,the three existing emotion dictionaries are integrated.In addition,a network emotion dictionary is constructed by combining emoticons and network neologisms.According to the characteristics of micro blog and depression,related words are also added to expand the dictionary.The three parts are integrated to construct a depression dictionary to extract local features of the textual data.(3)Construction of a prediction model for depression combining local and global features: A depression prediction model combining local and global features of the text is designed.The model uses capsule network to make up for the shortcomings of convolutional neural network,and uses depression dictionary to accurately find depression-related texts.The depression dictionary in the model is used to select local features in the text.In addition,capsule networks are used to learn the overall characteristics of the text.In the output layer of the model,two methods are used to fuse the local features and the global features to get the final result of depression prediction.