Mental Health Assessment Of Qnline Forum User Based On Text Mining

The Mental Health Forum is a service-oriented online community maintained bya psychologist,people can anonymously express their mental health problems and ask a counseling specialist or other users for help.In such forums,there are some users who are suffering or even self-harming at any time.Psychologists need to find and respond to such users as soon as possible to prevent them from harming themselves.However,thousands of posts are published daily in the forum,which makes it difficult for psychological consultants to find such users and respond in time.Therefore,it is very important to automatically evaluate the mental health of online forum users.This paper is based on the data published by CLPsych2017 shaed task,,from two aspects to build the model to assess the urgency(crisis: very urgent,red: urgent,amber: not urgent,green: no intervention required)of the mental health status of online forum users.:(1)Online forum user mental health automatic assessment framework F3 TMH based on multi-feature fusion.The framework adopts four features fusion strategies:greedy based F3TMH_G,voting based F3TMH_V,late fusion based F3TMH_L and denoising autoencoder based F3TMH_DA,which combines the behavior and attribute characteristics of posts(or their authors),liguishtic features,content features(including N-Grams features,topic features,word vector features),contextual features to construct a classification model.The experiment found that compared with other features,the word embedding feature performed better in automatically assessing the degree of mental health crisis of online forum users;the late fusion strategy F3TMH_L2 is more conducive to identifying users with higher mental health crisis(crisis and red)posts,and F3TMH_DA has an advantage for identifying relatively large amounts of data in the Flagged class(the union of non-green class)posts.(2)The CNN model for automatic assessment of mental health under the guidance of psychological knowledge.A convolutional neural network model LIWC-CNN based on psychological knowledge LIWC dictionary is proposed.Use the statistical features of the word frequency of the LIWC dictionary in different categories of posts,and use this to guide the convolutional neural network to extract posts(crisis and red)that are more conducive to identifying the interventions needed.The experimental results show that compared with other methods,the method used in this paper can improve the recognition effect of posts requiring urgent intervention at the expense of some users who can identify the accuracy of the posts that do not require intervention.It reflects the guiding role of psychological knowledge in the process of deep learning feature extraction.