Detecting Depressive Social Network Users Based On Text Minirung

With the development of society,the competition among people is more and more fiercer.People have long been in an semi mental health which is full of anxiety and panic,and mental illness break out in a short time.The development of social network such as micro-blog QQ and Wechat not only provides people a more convenient communication method,but also provides people a new outlet to express emotion.People can record their life-style in a real time and comment each other via social network,achieving the goal of relieving stress by expressing motion.At the same time,the development of social network also provides a innovative method to detect depressive users.We can apply the most advanced computer skill to analysis social network user's social text in order to detect their's mental healthy state.This paper's main task is to detect depressive user by analysis their social text using text mining technology.The main research work of this paper includes two parts.For the first part,we analyzed the feasibility of detecting depressive users using quasi-private social text.The related works usually took advantage of user behaviors data or text published on social network platforms such as Twitter and Micro-blog.There are few works use private data which come from the relatively private social network such as We Chat friends circle or QQ Zone to detect depressive users.Intuitively,this kind of quasi-private social network data can reflect user's mental health in a more exactly way.This paper mainly discusses the feasibility of detecting depressive users on quasi-private social network data,including training samples selection,feature quantification,detection model selection and the effects of detection model on the whole dataset,etc.The experimental result show that,to train an effective model and overcome the challenge of unbalance samples,we should firstly select almost the same amount of positive and negative samples with the highest and the lowest scores of self-report tests,which corresponding to the most depressive users and the most normal users.Secondly,the features should be quantified by Z-score standardized frequency,which is more powerful than the other two quantifying methods such as frequency or normalized frequency.Thirdly,the SGD classifier performs better than the other classifiers such as SVM on QQ zone data.Finally,compared to the QQ dataset,the model of detecting depressive users which is trained on Wechat data performs better on the whole dataset.In the other part,we explored the effect of detection model under different text features such as BOW feature,topical feature and word-to-vect feature.At present,most of the studies for detecting depressive users focus on the field of medicine and psychology.Due to lack of computer knowledge and limited human resources,the in-depth and comprehensive analysis of user data can not be conducted.Therefore,the detection of depressive social networks users can not be accurately and efficiently.In this paper,we extracted LIWC features by analyzing social text using LIWC dictionary which is often used to evaluate text data in psychology.We got BOW feature topic feature and word2 vec feature using text analysis technology which is often used in NLP,and built detection model using SGD classifier which is proved to be the best classifier.The results show that,compare to other features such as bag-of-words or word-to-vector,the detection model which is built on topical features performs better,where the F-measure of this model achieves 0.734,and the precision for depressive user detecting reached 0.813.At last,we aslo compared our result with related works,and the result show that our method for extracting text features can achive better than related works.