Enhancing Going Concern Prediction Models: Integrating Text Mining with Data Mining Approaches

The linguistic features embedded within business unit information play a crucial role in effectively conveying economic realities, a consideration increasingly recognized in accounting and behavioral finance research. This study endeavors to assess the predictive capacity of companies' going concern status by integrating structured and unstructured data, while also evaluating the impact of incorporating unstructured variables into traditional data mining models. Spanning from ۲۰۱۲ to ۲۰۲۱, the study encompasses a sample of ۵۴۰ company years listed on the Tehran Stock Exchange. Tone analysis of auditor reports was conducted using models by Mayew et al. (۲۰۱۵) and Visvanathan (۲۰۲۱), while MAXQDA ۲۰ text analysis software and the Loughran and McDonald (۲۰۱۵) dictionary facilitated data processing. Data analysis and hypothesis testing were performed using the logit regression model and the Vuong test. The findings support the first hypothesis, indicating that the text-based model yields a higher coefficient of determination compared to the data-based approach. Moreover, the second hypothesis reveals a significant discrepancy in the explanatory power between the data-based and integrated text-based models within companies

Keywords:

Auditor Report , Data mining , Going Concern , Sentiment Analysis , Text mining