Dual Attention based Suicide Risk Detection on Social Media

Despite some progress have been made in social media based suicide risk detection. However, visual information form post is often ignored when doing suicide risk detection. In this study, we propose a dual attention mechanism to capture the implicit correlation between text and image from the same post and utilize the visual information to improve the performance on social media based suicide risk detection. Experimental results on 5,000 Sina Weibo users show that with dual attention mechanism, our suicide risk detection model (DAM) can obtain 90% high accuracy.