Group 1 - The analysis of financial fraud samples indicates that the communications services industry has the highest proportion of fraud cases, while the financial and utilities sectors have the lowest [3][18]. - From 2010 to 2018, the number and proportion of fraudulent companies increased from 7.5% to 17.9%, but there was a decline after 2019, reflecting stricter regulations following the revision of the Securities Law [3][18]. - Approximately 58.3% of fraudulent activities are exposed or terminated within 1-2 years, suggesting that systemic fraud is difficult to maintain over the long term [3][18]. Group 2 - A feature pool was constructed based on structured financial data from listed companies, retaining 100 indicators from an initial 378, which included 5,483 fraud samples and 42,046 control samples [3][26][28]. - The feature pool covers eight dimensions, including solvency, operational ability, profitability, development ability, ratio structure, risk level, cash flow analysis, and per-share indicators [26][28]. - The model's performance improved with the inclusion of sentiment tone factors, enhancing recall rates and reducing Type II errors, indicating the importance of sentiment analysis in identifying financial fraud [3][48][56]. Group 3 - The sentiment tone factor was constructed using the DeepSeek R1 model to analyze financial report texts, revealing that companies involved in fraud generally have lower sentiment scores compared to normal companies [3][39][40]. - Emotional contradictions, vague expressions, and industry anomalies are identified as risk signals through sentiment analysis, which can indicate potential fraud [3][39][40]. - The sentiment analysis captures non-structured risk clues, helping to identify discrepancies between positive financial data and negative sentiment [3][39][40]. Group 4 - The report highlights that the communications services, energy, and financial sectors were the most affected by fraud from 2010 to 2018, with the communications services sector's fraud rate increasing significantly during this period [22]. - Information disclosure violations have become the mainstream method of fraud, with significant omissions and false records being the most common types of violations [22]. - The report notes that the real estate sector maintained a high fraud rate of 15.5% in 2022, indicating ongoing issues with financial misrepresentation in the industry [22].
基于财报文本的情感语调的分析:DeepSeek辅助识别财务瑕疵