Predicting Future Alleviation of Mental Illness in Social Media: An Empathy-Based Social Network Perspective

Yibo Chai, Fengyang Wu, Rui Sun, Zhongliang Zhang, Jie Bao, Runxin Ma, Qizhou Peng, Danqin Wu, Yexing Wan, Keyu Li

Abstract: Numerous studies have shown that users' posts on social media can explicitly or implicitly reflect various human psychological characteristics. Through mining these data, predictive models can be built to forecast and analyze potential mental illness, which can facilitate therapeutic decision making and offer the best hope for early interventions and treatments. However, most existing approaches face severe information loss and ignore the time dynamics of user behaviour. To fill the research gaps, we use time-aware social networks to combine various information sources in social media posts. Also, machine learning detectors are trained to automatically identify empathic interactions, filter out irrelevant information and construct empathy-based networks. Finally, we devise a hybrid deep learning algorithm to learn embeddings from the dynamic feature-rich networks and predict future alleviation of mental illness. Compared with strong baselines, our approach achieves the best-performing results with efficient computation speed.