Deep learning model with multi-feature fusion and label association for suicide detection

Suicide can cause serious harm to individuals, families, and society, and it has become a global social problem. Personal suicide ideation is concealed, and it is difficult to be accurately identified with traditional methods such as questionnaires and clinical diagnosis. With the development of the Internet, people are increasingly inclined to express their suicidal ideation on social media, where we can identify individuals with suicidal ideation. In this paper, we construct a Chinese social media suicide detection dataset, and extract the dictionary information of the posts, the user’s post time and social information. Then, we fuse the above features with deep learning methods, combine with our proposed label association mechanism, and raise a Text Convolutional Neural Network with Multi-Feature and Label Association (TCNN-MF-LA) model. Experiments show that the proposed model performs better than previous models. We also select some users in the dataset and analyze their posts to further clarify the effectiveness of the model. This work could help to enhance the identification of highest risk population groups and to avoid potentially preventable suicides.