Suicide ideation detection from online social media: A multi-modal feature based technique

Abstract

According to an estimate of World Health Organization, each year approximately 700,000 people die by suicide, with many more contemplating suicide. Early detection of suicidal ideation and proper treatment are two of the most effective techniques to preventing suicide attempts. People who are depressed or suicidal are increasingly using social media to express themselves. The main aim of this research is to provide early detection of suicide ideation by evaluating online social media. A well-labeled dataset of suicide thoughts was created on Reddit and Twitter and six feature groups were identified that included not just clinical suicidal symptoms but also online behaviors on social media. A multimodal model is proposed using these feature groups for identifying suicidal thoughts on social media. An accuracy of 87% was obtained using the Logistic regression classifier which outperforms other baselines. According to the study, effective feature selection and combination aids in obtaining greater performance.

Keywords

• Social media
• Twitter
• Suicide ideation
• Machine learning
• Natural language processing