Depression Risk Prediction for Chinese Microblogs via Deep-Learning Methods: Content Analysis (Preprint)

• DOI: 10.2196/preprints.17958
• Publication Date: 2020-02-06T21:03:03Z
• Authors: Xiaofeng Wang; Shuai Chen; Tao Li; Wanting Li; Yejie Zhou; Jie Zheng; Qingcai Chen; Jun Yan; Buzhou Tang
• Keywords: Microblogging; Depression; Shame; Benchmark (surveying)

<sec> <title>BACKGROUND</title> Depression is a serious personal and public mental health problem. Self-reporting the main method used to diagnose depression determine severity of depression. However, it not easy discover patients with owing feelings shame in disclosing or discussing their conditions others. Moreover, self-reporting time-consuming, usually leads missing certain number cases. Therefore, automatic discovery from other sources such as social media has been attracting increasing attention. Social media, one most important daily communication systems, connects large quantities people, including individuals depression, provides channel In this study, we investigated deep-learning methods for risk prediction using data Chinese microblogs, which have potential more trace conditions. </sec> <title>OBJECTIVE</title> The aim study was explore state-of-the-art on microblogs. <title>METHODS</title> Deep-learning pretrained language representation models, bidirectional encoder representations transformers (BERT), robustly optimized BERT pretraining approach (RoBERTa), generalized autoregressive understanding (XLNET), were prediction, compared previous manually annotated benchmark dataset. assessed at four levels 0 3, where 0, 1, 2, 3 denote no inclination, mild, moderate, severe risk, respectively. dataset collected microblog Weibo. We also different models two settings: (1) publicly released (2) further large-scale unlabeled Precision, recall, F1 scores performance evaluation measures. <title>RESULTS</title> Among three methods, achieved best microaveraged score 0.856. RoBERTa macroaveraged 0.424 represents new result demonstrated improvement over models. <title>CONCLUSIONS</title> applied automatically predict experimental results showed that performed better than greater