Article: Enterprise financial fraud detection based on double-layer knowledge graph and graph-driven algorithms Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.85 pp.47 - 69 Abstract: Financial statement fraud often hides behind locally plausible numbers while breaking cross-statement coherence, which weakens traditional ratio screens and single-source detectors. To address this challenge, a dual-layer knowledge graph and graph-driven fraud detection framework is proposed. In the framework, first, accounting semantics and consistency constraints are organised as a schema layer and linked to firm-year observations as an instance layer. Then, layer-coupled graph learning aligns observations with accounting meaning and extracts compact evidence subgraphs for audit review. Finally, a generative consistency module estimates coherent reporting profiles to expose distributed manipulation through residual inconsistency. Experimental results show that the proposed method achieves an average precision of 0.74 and improves low-false-alarm recall to 0.60, outperforming representative baselines by up to 0.14 in average precision. The framework provides reliable risk ranking and practical interpretability for audit-oriented deployment. Inderscience Publishers - linking academia, business and industry through research

Title: Enterprise financial fraud detection based on double-layer knowledge graph and graph-driven algorithms

Authors: Tiantian Wang

Addresses: Faculty of New Commercial Science, Anhui Sanlian University, Hefei, 230031, China

Abstract: Financial statement fraud often hides behind locally plausible numbers while breaking cross-statement coherence, which weakens traditional ratio screens and single-source detectors. To address this challenge, a dual-layer knowledge graph and graph-driven fraud detection framework is proposed. In the framework, first, accounting semantics and consistency constraints are organised as a schema layer and linked to firm-year observations as an instance layer. Then, layer-coupled graph learning aligns observations with accounting meaning and extracts compact evidence subgraphs for audit review. Finally, a generative consistency module estimates coherent reporting profiles to expose distributed manipulation through residual inconsistency. Experimental results show that the proposed method achieves an average precision of 0.74 and improves low-false-alarm recall to 0.60, outperforming representative baselines by up to 0.14 in average precision. The framework provides reliable risk ranking and practical interpretability for audit-oriented deployment.

Keywords: financial statement fraud; dual-layer knowledge graph; heterogeneous graph learning; evidence subgraph extraction.

DOI: 10.1504/IJICT.2026.155484

International Journal of Information and Communication Technology, 2026 Vol.27 No.85, pp.47 - 69

Received: 03 Mar 2026
Accepted: 08 May 2026
Published online: 03 Aug 2026 *

Title: Enterprise financial fraud detection based on double-layer knowledge graph and graph-driven algorithms

Keep up-to-date