The increased use of computer technology in education has led to a corresponding increase in computerized testing in schools (Dooey, 2008). The COVID-19 pandemic has sped up the transition to online testing (Adedoyin & Soykan, 2020). As a result of the pandemic, students are displaying more signs of anxiety and depression, due in part to the lack of social interaction, a potentially strained home environment, feeling on their own to learn, and losing track of assignments (Cao, et al., 2020; Noble, Hurley, & Macklin, 2020). Teachers may not be able to see their students due to social distancing and students may be reluctant to reveal issues through a virtual format (Son et al., 2020). Identifying indicators of mental disorders can be difficult as indicators manifest in different ways (Muris et al., 2001). Identifying these signs early can have a significant impact on a child’s life (McLoone, Hudson & Rapee, 2006). A method for detecting potential indicators in students’ writing may help educators identify early signs of risky behavior. There has not yet been reported a method of computerized analysis for detecting indicators of risky behavior or negative sentiment through students’ writing. In this study, I investigate using topic modeling (Blei, 2012) as a method of detecting potential indictors of negative sentiments. The objective of this study is not to diagnose students of disorders through their essays. Rather, it is to explore the development of an internal tool that could be used to assist educators in flagging potentially at-risk students. Topic modeling was used to analyze students’ written responses to standardized test questions for detection of negative sentiment as potential indicators of future risky behavior. Results indicate topic models can identify negative sentiment and risky behavior. Additional validation measures such as sentiment analysis and human raters were used to confirm topic modeling results.
