Emotion Expression Analysis And Prediction Of Depressed Patients Based On Emotion Lexicon

Emotion analysis based on social media is a hot topic of research.However.previous studies mainly focus on the dichotomy of positive and negative,with less emotion analysis on finer granularity.At the same time,previous emotion lexicon often confuse the difference between emotion description and emotion expression.Therefore,this study hopes to build a more granular emotion expression lexicon.Previous studies have found that depressed patients express more negative emotions on social media.and emotion characteristics are an eff-ective feature in predicting depression.Based on the multi-classification emotion lexicon,this paper hopes to further study the difference of emotion expression between depressed patients and ordinary users under the condition of fine granularity,and to explore the effect of fine granularity emotion characteristics in predicting depressed patients.Therefore,this study hopes to build a more granular emotion expression lexiconThis study explores the emotion expression of users through the method of vocabulary matching.Firstly.an emotion expression lexicon was constructed and compared with the existing emotion lexicon.It was found that the coverage and accuracy of the lexicon were significantly improved.Then,based on the lexicon and matching rules,the differences in emotion expression between depressed patients and ordinary users were analyzed.Finally,the emotion scores of each dimension were used as the characteristics to construct the prediction model for identifying depression.In study 1,first,three graduate students from the department of psychology conducted emotion evaluation on the microblog posts,and the discussion led to the classification of emotion dimensions.Then,emotion seed words of different dimensions are extracted and expanded to obtain the emotion classification lexicon.Then,by comparing the effect of the lexicon with that of the predecessors,it is found that the effect of the lexicon is better than that of other lexicon.Finally,10 psychology students were recruited to score the words in the lexicon in terms of pleasure,arousal and dominance by 9 points,so as to obtain the lexicon of emotion dimensionIn study 2,first,microblog posts was obtained for patients with depression and ordinary users.Then,each microblog of the user is labeled with emotion through the method of word matching,and the percentage of emotion microblog in the total microblog and the mean value of pleasure,arousal and dominance of emotion vocabulary are calculated respectively.Finally,the difference between the depression group and ordinary users was analyzed,and it was found that the depression group expressed more sadness,anxiety and disgust,and the overall emotion pleasure and dominance were lower.In study 3,first,the emotion scores of each dimension of patients with depression and ordinary users in study 2 were selected as the characteristic input.Multiple algorithms(random forest,naive bayes,logistic regression and gradient rise)were used to construct the prediction model,and the optimal fl score of the model reached 86.6%.The model's generalization capability is then further verified using data that is not part of the built model.Therefore,the following conclusions can be drawn from this study:(1)The multi-category emotion expression lexicon constructed has better effect than other lexicon and can be used as a tool for emotion expression analysis.(2)Depressed patients express more sadness,anxiety and disgust than ordinary users on weibo.(3)Depressed patients express lower degree of emotional pleasure and dominance than ordinary users on weibo.(4)It is effective to use emotion score as a feature to construct a depression prediction model,among which sadness score is the most effective feature,and the model has a good generalization ability.