According to the World Health Organization (WHO), approximately 1.4 million individuals died by suicide in 2022. This means that one person dies by suicide every 20 seconds. Globally, suicide ranks as the 10th leading cause of death, while it ranks second for young people aged 15-29. In the year 2022, it was estimated that about 10.5 million suicide attempts occurred. The WHO suggests that alongside each completed suicide, there are many individuals who make attempts. Today, social media is a place where people share their feelings, such as happiness, sadness, anger, and love. This helps us understand how they are thinking or what they might do. This study takes advantage of this opportunity and focuses on developing an automated tool to find if someone may be thinking about harming themselves. It is developed based on the Suicidal-Electra model. We collected datasets of social media posts, processed them, and used them to train and fine-tune the model. Upon evaluating the refined model with a testing dataset, we consistently observed outstanding results. The model demonstrated an impressive accuracy rate of 93% and a commendable F1 score of 0.93. Additionally, we developed an API enabling seamless integration with third-party platforms, enhancing its potential for implementation to address the growing concern of rising suicide rates.

翻译：据世界卫生组织统计，2022年全球约140万人死于自杀，相当于每20秒就有一人自杀身亡。在全球范围内，自杀已成为第十大死因，在15-29岁青年群体中则高居第二位。据估计，2022年全球发生了约1050万次自杀未遂事件。世卫组织指出，每一起自杀死亡背后都有更多未遂案例。如今，社交媒体已成为人们表达喜怒哀乐等情感的平台，这为理解个体思维状态和潜在行为提供了可能。本研究借此机遇，致力于开发一种基于Suicidal-Electra模型的自动化工具，用于检测个体是否存在自残倾向。我们收集并处理了社交媒体帖文数据集，通过训练与微调优化模型性能。经测试数据集评估，该模型始终表现优异：准确率达93%，F1分数达0.93。此外，我们开发的应用程序编程接口（API）可实现与第三方平台的无缝集成，为应对日益严峻的自杀率问题提供了切实可行的解决方案。