Depression Detection on Twitter Text Based on Negative Emotion Score and Level of Depressed

Abstract

This essay presents a method for detecting depression level detection based on the negative emotion score of social media text. The score is calculated by the NRC Word-Emotion Association Lexicon (EmoLex). The dataset will first be processed by the EmoLex model and then based on certain scores given by the model, three levels of depression will be generated for further classification. The advantage of this model is that the detected depression is classified into three levels, which can help people take different levels of response. The experiment is performed in Multi Labeled Depression Corpus of 60,000 English tweets. Three models are trained in this experiment: Random Forest and Logistic Regression. The result shows that Random Forest achieves 65.9% accuracy; Logistic Regression achieve 72.2% accuracy.