Depression detection on online social network with multivariate time series feature of user depressive symptoms

In recent years, depression has attracted worldwide attention because of its prevalence and great risk for suicide. Existing studies have confirmed the feasibility of depression detection on online social networks. Most existing researches extract the overall features of users during a specific period, which cannot reflect the dynamic variation of depression. Besides, the methods proposed in these studies are often lack in interpretability and fail to establish the correlation between features and depressive symptoms in clinical. To address these problems, we propose a novel framework for depression detection based on multivariate time series feature of user depressive symptoms. Firstly, we construct and publish a well-labeled dataset collecting from the most popular Chinese social network platform Sina Weibo. To the best of our knowledge, it is the first large-scale depression dataset with complete collection of user tweeting histories, which includes 3,711 depressed users and 19,526 non-depressed users. Then, we propose a feature extraction method that reveals user depression symptoms variation in the form of multivariate time series. Moreover, we explore the various influencing factors to the performance of our proposed framework. In addition, we also explore the contributions of features to classification as well as their interpretability and conduct feature ablations on them. The experimental results show that our proposed method is effective and the extracted multivariate time series feature can well characterize the depressive state variation of users. Finally, we analyze the shortcomings and challenges of this study. Our research work also provides methods and ideas for tracking and visualizing the development of depression among online social network users.