Article: Edge-empowered hierarchical attention network for accurate English text source tracing Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.79 pp.70 - 93 Abstract: Text provenance aims to determine the authorship of an anonymous text, which is crucial in the field of digital forensics and academic integrity. Existing models are difficult to deal with hierarchical style features of long documents, and cannot run in real-time on edge devices. This paper proposes edge-empowered hierarchical attention network, which integrates the cognitive load theory, extracts writing habits through the word-sentence-paragraph-full text layer-by-layer attention mechanism, and designs constraints to distinguish scene differences without confusing author identities. Finally, the model is compressed to the Edge end by pruning and distillation. On the PAN 2020 dataset, the traceability accuracy of the proposed framework reaches 0.923, which is 4.2% higher than the current best method. The edge-side inference takes only 12.3 Ms. This paper quantifies cognitive load theory as a computable constraint for the first time, and provides a high-precision solution for real-time text provenance at the edge. Inderscience Publishers - linking academia, business and industry through research

Title: Edge-empowered hierarchical attention network for accurate English text source tracing

Authors: Hao Tang

Addresses: School of Foreign Studies, Xiangnan University, Chenzhou, 423000, China

Abstract: Text provenance aims to determine the authorship of an anonymous text, which is crucial in the field of digital forensics and academic integrity. Existing models are difficult to deal with hierarchical style features of long documents, and cannot run in real-time on edge devices. This paper proposes edge-empowered hierarchical attention network, which integrates the cognitive load theory, extracts writing habits through the word-sentence-paragraph-full text layer-by-layer attention mechanism, and designs constraints to distinguish scene differences without confusing author identities. Finally, the model is compressed to the Edge end by pruning and distillation. On the PAN 2020 dataset, the traceability accuracy of the proposed framework reaches 0.923, which is 4.2% higher than the current best method. The edge-side inference takes only 12.3 Ms. This paper quantifies cognitive load theory as a computable constraint for the first time, and provides a high-precision solution for real-time text provenance at the edge.

Keywords: text provenance; hierarchical attention; cognitive load theory; edge computing; authorship attribution.

DOI: 10.1504/IJICT.2026.155059

International Journal of Information and Communication Technology, 2026 Vol.27 No.79, pp.70 - 93

Received: 10 Apr 2026
Accepted: 16 May 2026
Published online: 24 Jul 2026 *

Title: Edge-empowered hierarchical attention network for accurate English text source tracing

Keep up-to-date