Financial fraud detection using the related-party transaction knowledge graph
Xuting Mao, Hao Sun, Xiaoqian Zhu, Jianping Li
Abstract
Financial fraud detection has gained constant attention from researchers, practitioners, and regulators. Because of its concealment and ease of manipulation, related-party transactions (RPTs) among firms have become a usual way to implement financial fraud. However, the traditional quantitative analysis methods regard each firm as an independent individual, failing to mine the intricate relation and transactions among related parties. A knowledge graph can mine valuable hidden knowledge from large-scale associated data as a new form of knowledge representation. Therefore, in this paper, the RPT knowledge graph is utilized to detect financial fraud, where the feature of the transaction’s scale and category is obtained. The experiment on the Chinese listed companies from 2000 to 2019 shows that these features enhance financial fraud detection performance, suggesting that type, size, and frequency of RPTs may imply fraud. More details, the feature importance indicates that regulators should pay more attention to the loan-based RPTs and the total number of RPTs.