Study on Mental Disorder Detection via Social Media Mining

Traditional mental health studies rely on information collected through personal contact with professional healthcare specialists. Recent work shows the utility of social media data for studying mental disorder. It is supported by the massive usage of social media and the disclosure that social media is a pool of emotion. Study on social media data could potentially complement the traditional technique in its ability to provide natural measurements. We build a corpus of self-declared mental illness diagnosis on Twitter using a source of publicly-available data. This study implements computational linguistic process with linguistic and emotion feature to model the rate of depression in social media data. We propose the features of SenticNet's four dimensions emotional state of the mind, self-reference, and mental disorder wordcount. The results are shown using a rule-based system to determine the level of depression based on language. We found 1,733 typical words from the depression diagnosed group. This finding is based on the match of Wordnet, SenticNet, Vader, TextBlob, and has been evaluated by synonyms checking. Our proposed method has been successfully identified and then categorized 8105 tweets into 3 levels of depression, 1028 tweets are categorized as high, 1073 moderate, and 1605 low.