An Initial Study of Depression Detection on Mandarin Textual through BERT Model

Abstract

Depression is one of the most common mental disorders nowadays. Cases of depression is increasing significantly in Malaysia during Covid-19 pandemic. In the era of advancement of Internet technology, number of social media user is growing exponentially and became part of human lifestyle. Social media has provided a platform to their user to share their thought and feelings effortlessly. Previous studies demonstrated that the possibility and capability of artificial intelligence technology on analyzing texts on social media for detecting depression tendency. However, most of the study are conducted on English textual content. Mandarin is ranked second popular spoken language in the world, thus it is worth to explore depression detection technique on Mandarin textual content. In this study, BERT model is proposed for conducting depression detection on social media. Mandarin text data is targeted as there are less studies exploring to depression detection on Mandarin text. In addition, it is also worth to find out the capability and performance of pre-trained BERT model on this particular tasks especially on Mandarin language. This is an initial study and there are more works yet to be done. This paper will focus on the dataset acquisition and analysis.