ÃÊ·Ï
º» ¿¬±¸´Â Çѱ¹ÀÎ Àç³°æÇèÀÚÀÇ ¿Ü»ó ÈÄ ½ºÆ®·¹½º Àå¾Ö(PTSD) À§Ç豺À» ¼±º°Çϱâ À§ÇÑ Çѱ¹¾î Æ¯È ÀΰøÁö´É(AI) ¸ðµ¨ °³¹ßÀ» ¸ñÇ¥·Î ÇÑ´Ù. ±âÁ¸ ´ë¸é Æò°¡ ¹æ½ÄÀÇ Á¢±Ù¼ºÇѰ踦 ±Øº¹ÇϰíÀÚ, 1,249¸íÀÇ Àç³°æÇèÀڷκÎÅÍ ¼öÁýÇÑ À½¼º ÀÎÅÍºä ±â¹ÝÀÇ ÅØ½ºÆ® µ¥ÀÌÅͼÂÀ» ±¹³» ÃÖÃÊ·Î ±¸ÃàÇÏ¿´´Ù. Çѱ¹ÆÇ PTSD üũ¸®½ºÆ®(K-PCL-5)¸¦ ±âÁØÀ¸·Î µ¥ÀÌÅ͸¦ Áõ»ó±º°ú Á¤»ó±ºÀ¸·Î ºÐ·ùÇϰí, KoBERT, KoELECTRA, KoRoBERTa ¸ðµ¨À» ¹Ì¼¼Á¶Á¤ÇÏ¿© ¼º´ÉÀ» Æò°¡Çß´Ù. ½ÇÇè °á°ú, ¹ßÈÀÚÀÇ ¹è°æ Á¤º¸¿Í ´ëÈ ¸Æ¶ôÀ» Æ÷ÇÔÇϰí, °³ÀÎÀÇÀüü ¹ßȸ¦ ÇϳªÀÇ ¼»ç·Î ºÐ¼®ÇÏ´Â ¹æ½ÄÀÌ °¡Àå È¿°úÀûÀ̾ú´Ù. ÃÖ»óÀ§ ¸ðµ¨ÀºF1-Score 0.838, Áõ»ó±º ÀçÇöÀ² 0.75ÀÇ °æÀï·Â ÀÖ´Â ¼º´ÉÀ» º¸¿©, ÇâÈÄ ºñ´ë¸é ¿Â¶óÀÎPTSD Æò°¡ ¼ºñ½ºÀÇ ÇÙ½É ±â¼ú·Î¼ÀÇ ½Ç¿ëÀû °¡´É¼ºÀ» ÀÔÁõÇÏ¿´´Ù.
This study aims to develop a Korean-specific AI model to screen for Post-Traumatic Stress Disorder (PTSD) risk groups among Korean disaster survivors. To overcome the accessibility limitations of traditional face-to-face assessments, a novel text dataset was constructed from voice interviews with 1,249 survivors. Using the Korean PTSD Checklist (K-PCL-5) for classification, the performance of several fine-tuned language models was evaluated. The results indicated that the most effective approach involved analyzing an individual's entire narrative combined with their background and conversational context. The top-performing model demonstrated competitive performance with an F1-Score of 0.838 and a symptom group recall of 0.75, proving its practical potential as a core technology for future remote online PTSD assessment services.







