Suicide behavior is an enormous challenge in our society. More than 700,000 people died for this reason every year (OMS, 2023). This emphasizes the need of developing new strategies that help to identify risk factors to arise prevention strategies. This study emerges with the objective, on the one hand of applying new methods for detecting suicide behavior based on integration of Natural Language Processing (NLP) techniques and on a second hand, explore risk factors associated. The study explores the integration of NLP techniques with traditional clinical assessments to improve suicide risk evaluation. We analyzed qualitative responses from 1,443 participants to the Columbia-Suicide Severity Rating Scale (C-SSRS) using advanced BERT-based models. Our approach combined NLP analysis with logistic regression to identify psychological factors associated with various aspects of suicidal ideation and behavior. The results show that the NLP models demonstrated high accuracy in classifying different types of suicidal ideation and preparatory behaviors, with an overall accuracy of 83% and particularly effective in identifying passive ideation (94% precision). Significant risk factors were identified, including cognitive instability, death anxiety, and substance use. Cognitive instability and perseverance were associated with a higher probability of active ideation, while death anxiety showed an inverse relationship. Medication use was strongly linked to preparatory storage behaviors for suicide. The study revealed gender and age differences in suicide attempt prevalence, with a higher incidence in women and young adults. The integration of NLP models in clinical assessment allowed for the identification of discrepancies between descriptive responses and structured data, offering a more comprehensive understanding of patients' emotional and cognitive states. As a conclusion, the findings highlight the potential of combining computational techniques with clinical expertise to enhance suicide risk assessment. This approach offers promising avenues for developing more effective strategies for early detection and intervention in suicidal behavior, emphasizing the importance of personalized interventions based on individual psychological profiles.

Suicide risk assessment, natural language processing, machine learning, C-SSRS, mixed-methods research

Applied Psychology Around the World | Volume 7, Issue 2