In the field of public health, suicide risk prediction is a central and urgent problem. Existing researches mainly focus on user’s current post but overlook historical post. In light of the psychological characteristics, we argue that it is valuable to consider users’ historical post in addition to current post for predicting suicide risk. Based on this rationale, we propose a deep learning-based suicide risk prediction framework - Dynamic Historical Information based Suicide Risk Prediction (DHISRP) - by considering the user’s current post content and historical post content. To capture the dynamic and complicated information of historical post, we design a unit based on long short-term memory (LSTM), named RNLSTM. We also conduct experiments to compare with the benchmark model to prove the effectiveness of our model, and perform ablation experiments to verify the significance of each component in the prediction framework in this study.


Paper Number 1438; Track Design; Complete Paper


When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.