Identifying Suicidal Ideation Through Automatic Extraction of Emotional Traces in Suicide Notes

Suicide is a critical mental health issue and one of the leading causes of death worldwide. Several studies in suicidology claim that emotional dysregulation is a contributing factor in triggering suicidal behavior. In today’s digital age, individuals often express suicidal ideation on social media platforms to seek help, empathy, or advice. The primary objective of this study is to classify suicide notes based on their emotional content using machine and deep learning algorithms. We propose an innovative approach to automatically identify emotional changes that occur through the ideas expressed in a suicide note, leveraging these shifts as a key indicator of suicidal ideation. The goal is to automatically identify the correlation between latent emotional states in suicide notes and their correct classification (suicide or not). We employed a long short–term memory neural network that captures emotional distributions to address several binary classification scenarios. Results demonstrated a performance with an f-measure exceeding 80% in all suicide note classification scenarios. Our analysis revealed that the vocabulary used in suicide notes collected from social media often differs from that found in notes written by individuals who have actually completed suicide. Additionally, we observed that suicide attempters typically do not directly express suicidal ideation, instead using non-death-related vocabulary, which presents a greater challenge in identifying genuine suicidal thoughts.