Artificial Intelligence Detection of Earnings Management in Reporting Systems Based on Natural Language Processing Technology
Year: 2026
Type: conference-paper
Abstract: To address the problem of insufficient utilization of unstructured information in traditional earnings management identification methods, this paper introduces natural language processing technology to construct an intelligent identification model that integrates textual semantic features and financial data. By semantic vectorization, sentiment feature extraction, and multi-dimensional feature fusion of corporate annual report texts, combined with a deep learning model, earnings management behavior is identified. Experimental results show that the proposed method outperforms the comparison model in recognition performance, achieving an accuracy of 0.861 and an F1 score of 0.839, an improvement of approximately 9.8% compared to traditional methods. Meanwhile, feature ablation experiments show that the model performance improves from 0.781 to 0.861 after introducing semantic vectors, verifying the crucial role of textual information in the identification process. Furthermore, the model performs stably under different training set proportions, with accuracy remaining between 0.821 and 0.861, demonstrating good generalization ability. The research results indicate that this method can effectively improve the accuracy of earnings management identification, providing technical support for intelligent auditing and supervision.
Author: Yanrong Mou
Institution: Wuhan Qingchuan University
Language: English
Subfield: Accounting
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7201849030