A Corporate Financial Statement Fraud Model Based on Extracting Correlated Topic Features

Fraud in corporate financial statements will cause significant losses to stakeholders, companies, and the country's economy. However, how do you ensure that individuals and companies are protected from losses? In response to this problem, this study proposes a weighted correlated topic model to establish a financial fraud detection system, providing an effective and applicable solution for detecting financial statement fraud, and the system can detect financial fraud on a large number of textual documents. This study evaluates the best topic numbers based on the coherence metrics to extract concise features by using the proposed weighted CTM. To evaluate the efficacy of the proposed detection model, we collected management discussion and analysis (MDA) documents of the US to conduct experiments to compare the proposed model with the listing model. The results show that the combination of the proposed NPMI+CTM method with random forest has the best performance.