Title: Multimodal legal information extraction based on gated graph neural network

Authors: Lizhi Yuan

Addresses: School of Political Science and Law, Hubei University of Arts and Science, Xiangyang, Hubei 441000, China

Abstract: To augment the precision and reduce the extraction duration of legal information, a multimodal extraction technique utilising a gated graph neural network is introduced. Initially, multimodal legal data is subjected to binary transformation and dimensionality reduction processes. Following this, intricate textual and visual details from legal documents are captured by a bidirectional gated recurrent unit. An attention mechanism is employed to assess the correlation between legal text and images, with single-mode relationships being modelled through an intra-modal attention module, resulting in a unified multimodal representation. By integrating legal text and image features through a gated neural network enhanced with dropout, efficient multimodal extraction is achieved. Experimental results demonstrate that the accuracy and efficiency of multimodal legal information extraction are significantly improved by this method, with an extraction precision consistently maintained above 90%.

Keywords: gated graph neural network; multimodal method; legal information extraction; text features; image features.

DOI: 10.1504/IJCAT.2026.154043

International Journal of Computer Applications in Technology, 2026 Vol.78 No.4, pp.273 - 280

Received: 07 Nov 2024
Accepted: 22 Apr 2025
Published online: 10 Jun 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article