Learning Depression Patterns From MyPersonality and Reddit
Millions of Americans suffer from mental health problems caused by negative sentiment such as depression and neuroticism. Nevertheless, traditional clinical diagnosis is not preferred as a large-scale approach due to the high expenditures and subjective judgement. In this paper, we conduct exploratory data analysis on three datasets from MyPersonality and Reddit, train and compare different predictive models for depression and neuroticism with various features. Our experimental results indicate promising feasibility to identify such mental health problems using natural language processing techniques. Moreover, we discuss the differences among χ 2 -Test, PMI, LDA, between single and multiple feature selections, and between SVM and SLDA. Thus, we accomplish and overfulfill all of our proposed goals. Eventually, we propose several directions for future research.
