基于文本大数据分析的年报舞弊识别研究

Research on Annual Report Fraud Identification Based on Text Big Data Analysis

上市公司通过年报对外公布财务状况,部分企业为了自身利益,采用伪造、隐瞒的方式选择性披露信息,从而出现财务报告舞弊的现象。监管部门为了治理年报舞弊,采取多种策略规避会计行为、提高会计信息质量、促进市场与投资者良性交互。经研究,年报除去部分经营性指标,存在大量文本信息披露,该内容信息量巨大且隐蔽性强,因此提高财报文本的自动化甄别能力具有重要意义。本文基于BERT获得了86.57%的分类精度,具有一定的判别有效性,本文对测评结果进行了详细分析。

Listed companies publish their financial status through annual reports,and some enterprises selectively disclose information in the form of forgery and concealment for their own interests,which leads to financial reporting fraud.In order to control the fraud in the annual report,the regulatory department adopts various strategies to avoid accounting behavior,improve the quality of accounting information,and promote the benign interaction between the market and investors.After research,the annual report has a large amount of text information disclosure except for some operational indicators,which is huge and hidden.Therefore,it is of great significance to improve the automatic screening ability of the financial report text.This paper obtains 86.57%classification accuracy based on BERT,which has certain discrimination effectiveness.This paper analyzes the evaluation results in detail.