An Intelligent Authenticity Identification Model for Accounting Invoices Through Transformer-Based Bidirectional Semantic Representation Approach

With the rapid development of accounting informatization, the authenticity and accuracy of accounting documents have become crucial to the financial health of enterprises. However, traditional accounting and auditing methods are increasingly unable to meet the demands for efficient identification in the face of complex and evolving fraud techniques. To address this challenge, this study proposes an intelligent authenticity identification model for accounting invoices using a Transformer-based bidirectional semantic representation approach. By capturing the contextual information of accounting documents, this method constructs an efficient semantic representation model that enables a deep understanding of document content. The paper employs the self-attention mechanism and positional encoding techniques within the Transformer architecture to effectively extract key information from documents, while bidirectional training strategies are implemented to enhance the model’s understanding of semantics in both directions. In experiments, we use a substantial amount of real accounting document data to train and test the model. The results demonstrate that the model performs well in identifying fraudulent accounting documents, with a significant improvement in accuracy and strong generalization ability. Compared to other traditional methods and existing advanced models, the proposed model shows notable advantages in accuracy, F1 score and robustness.