Article: Financial fraud recognition based on deep learning and textual feature Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.25 No.12 pp.1 - 15 Abstract: Financial fraud refers to the egregious breach of trust that uses improper means to distort accounting information, which negatively affects the company's operation. Intending to the issues of ignoring text features in existing research, a financial fraud recognition method based on deep learning and text features is designed. The method starts with preprocessing financial indicators and uses BiLSTM to extract sentiment features from the text of the management discussion and analysis (MD&A) chapter. Then the parallel residual network is used to select the financial indicator variables and textual sentiment variables in depth, the selected two variables are inputted into the dual-channel CNN for feature extraction, and feature enhancement and fusion are carried out using multi-head attention, and the recognition results are outputted through softmax. The experimental results show that the proposed model achieves better financial fraud identification, with the accuracy and AUC reaching 91.35% and 98.52%, respectively. Inderscience Publishers - linking academia, business and industry through research

Title: Financial fraud recognition based on deep learning and textual feature

Authors: Weiyi Chen

Addresses: Monitoring and Audit Department of the Financial Shared Center, National Energy Group Qinghai Electric Power Co., Ltd., Xining, 8180000, China

Abstract: Financial fraud refers to the egregious breach of trust that uses improper means to distort accounting information, which negatively affects the company's operation. Intending to the issues of ignoring text features in existing research, a financial fraud recognition method based on deep learning and text features is designed. The method starts with preprocessing financial indicators and uses BiLSTM to extract sentiment features from the text of the management discussion and analysis (MD&A) chapter. Then the parallel residual network is used to select the financial indicator variables and textual sentiment variables in depth, the selected two variables are inputted into the dual-channel CNN for feature extraction, and feature enhancement and fusion are carried out using multi-head attention, and the recognition results are outputted through softmax. The experimental results show that the proposed model achieves better financial fraud identification, with the accuracy and AUC reaching 91.35% and 98.52%, respectively.

Keywords: financial fraud identification; deep learning; DL; text feature; BiLSTM; CNN.

DOI: 10.1504/IJICT.2024.143633

International Journal of Information and Communication Technology, 2024 Vol.25 No.12, pp.1 - 15

Received: 27 Oct 2024
Accepted: 22 Nov 2024
Published online: 02 Jan 2025 *

Title: Financial fraud recognition based on deep learning and textual feature

Keep up-to-date