抑郁症是中国乃至全球医疗和卫生保健的沉重负担和工作重心。已有的心理健康研究方法主要依赖于比较费时费力的调查问卷和临床交流，整体对人口的普及性较低，这使得基于社交媒体的在线互助社群成为患者寻求社会支持的主要途径之一。患者在线留下的行为数据包含了其社交行为和语义特征，这些数据可以作为评估和预测患者抑郁严重程度的重要指标。然而，如何将这些高维数据整合一起进行分析还缺乏一个统一的建模机制。..本课题提出了一种新的基于张量分解的建模框架，用四维张量将用户的复杂在线动态行为进行统一建模，从而可以开发机器学习模型，基于真实历史数据对患者的行为和病状进行建模并针对个体做出预测。本课题计划通过临床医生的人工对在线行为数据进行抑郁严重程度标记，训练可以预测患者抑郁严重程度的预测模型，并计划通过临床病人的实验，进一步验证所开发的预测模型在实际治疗中帮助医生及时感知病人病情走向从而进行前瞻性干预的有效性。

Mental health is a major health issue worldwide. Existing mental health assessment approaches heavily rely on the labor-intensive surveys and clinical interactions, which lead to the low coverage and unreliable findings. The digital footprints of Web users left on social media present important mental health proxies with new opportunities to detect the risk factors of mental health in addition to traditional interventions/promotion programs. How to use this valuable multiaspect data source to identify the risk factors of mental health problems (like depression) using the social media users’ online behaviors still remains under-researched...This project aims to address this challenge by proposing data-driven research on characterizing the dialogues and behaviors as discussed on social media that are of value to depression assessment. We will develop a tensor-based model to assess the risk of depression for users using the multiaspect social, semantic, and temporal relations revealed by their behaviors on social media. The research outcomes will be validated using the ratings of depression generated by mental health counselors and self-assessment through real-world experiments. The proposed project could provide valuable data-driven decision support to help healthcare providers assess the risk factors of depression in population.