Detection of Suicide Risk in Korean Text Using Deep Learning

This study presents a model for detecting suicide ideation in Korean online and on SNS using deep learning. With the introduction of natural language learning and artificial intelligence technology, we were not only able to understand language but also read emotions inherent in text, making great progress. There has been an increasing number of studies using these technologies to detect the suicide risk of online users. In addition, as the industry commercializes related technologies, it is possible to respond by analyzing suicide-inducing information through online and SNS to detect users of high risk of suicide at an early stage. Although Korea has a high suicide rate, related research is still insufficient. CNN, GRU, and LSTM deep learning models were used for the experiment, and each learning set and test set consisted of 31,770 and 6,354 data. In the analysis results, the LSTM model showed the best performance, and showed a classification accuracy of 65% in articles with suicidal risk and 99.18% of classification accuracy in articles with non-suicidal risk. This study is the first study to show the possibility of detecting signs of suicide risk in Korean, and offers implications for academia and industry.