A Graph Mining Approach to Identify Financial Reporting Patterns: An Empirical Examination of Industry Classifications
提出一种基于可扩展商业报告语言（XBRL）财务分类的图挖掘方法，通过图相似度度量与谱聚类算法量化财务披露的相似性，能有效识别行业边界，并比传统行业分类方案（SIC、NAICS）具有更低的方差。
ABSTRACT This study proposes a quantitative method using the eXtensible Business Reporting Language financial accounting taxonomies to identify firms' common business characteristics and demonstrates that this graph mining approach can effectively identify industry boundaries. The premise of this method is based on the previous findings that financial accounts and the structural semantic information represented in financial statements reveal firms' general business operations and common characteristics if they have similar business models. Specifically, we introduce a graph similarity metric combined with spectral clustering algorithm to quantify the similarity of financial disclosures. Through industry classification comparison with the traditional classification schemes, the Standard Industrial Classification and the North American Industry Classification System, we show that the proposed method consistently clusters firms into their respective industries based on financial disclosures with significantly lower variance in a time‐varying fashion. This novel graph mining method provides an automated way for decision makers to identify common business operations as well as detecting potential financial fraud and uncovering accounting information misrepresentation.