Development and validation of an automated machine for self-injury assessment via young Koreans’ natural writings

Abstract

Self-injury is common in all countries, and 20% of South Korean youths experience self-injury. One of the barriers to assessment and treatment planning is the tendency of young self-injurers to conceal their identities. Following a new stream of research that uses online text data to assess psychological symptoms as they are described in online posts, this study developed a computerized machine that can analyze South Korean self-injurers’ writing in assessing their self-injury severity. Based on 16,645 online posts, Study 1 developed a machine called the Korean Self-Injurious Text Reviewer (K-SITR) using Latent Dirichlet Allocation topic modeling and machine learning. The K-SITR’s text-assessment results were statistically indistinguishable from those of professional counselors. Study 2 confirmed the validity of the K-SITR through a survey of 47 young Koreans who had experienced self-injury. Results showed that the K-SITR scores converged with participants’ self-injury frequency and duration and discriminated from other heterogenous factors. The K-SITR also had incremental validity over two popular self-injury questionnaires. This study provides a new measure that may reduce the tendency of young self-injurers to self-conceal compared to traditional direct-item questionnaires.