Assessing depression risk in Chinese microblogs: a corpus and machine learning methods

Abstract

Depression is a serious challenge for personal and public health. Tens of millions of people suffer from depression every year, but only a small part of them can receive professional treatment. Because people usually avoid disclosing or discussing mental health conditions, which are considered as stigma or taboo. Social network, such as the most popular Chinese microblogging website Sina Weibo, provides a channel for people, including depressed patients, to share their thoughts. We may find depressed emotions from microblogs. In this paper, we randomly collected a set of microblogs and manually annotated them with depression risk at level 0-3 (from mild to severe). On this corpus, we compared different machine learning methods for depression risk prediction, and provided benchmark results. The machine learning methods are support vector machine (SVM), convolutional neural network (CNN), long short-term memory network (LSTM) and bidirectional encoder representations from transformers (BERT).