Enhancing BiGRU with a KAN Block for Legal Document Classification and Summarization

2026-05-27

AI Summary

This study addresses the challenges posed by domain-specific linguistic complexity, multilingual diversity, long-range contextual dependencies, and class imbalance in low-resource multilingual legal documents. To tackle these issues, the authors propose a novel approach that integrates bidirectional gated recurrent units (BiGRU) with Kolmogorov–Arnold Networks (KAN)—marking the first application of KAN to legal text processing. The model combines an attention-enhanced GRU with a KAN-based head to jointly perform classification and abstractive summarization. Evaluated on a multilingual legal dataset comprising Bengali, English, and transliterated Bengali texts, the method achieves a classification accuracy of 67.96% (F1 = 0.65) and ROUGE-1/2/L scores of 0.38, 0.23, and 0.31 for summarization, respectively. Ablation studies demonstrate that the inclusion of the KAN module improves classification accuracy by over 10 percentage points, substantially outperforming baseline models.

Abstract

This study introduces a novel architecture of KAN-based BiGRU model for the task of classification and summarization of legal documents in a low-resource multilingual setup. In order to tackle problems associated with domain language, the usage of different languages, long dependencies within context, and class imbalance, we employ the dataset composed of legal documents from Bangladesh and taken from Manupatra, which include Bengali, English, and transliterated Bengali languages. Our classification task involves BiGRU model, along with Kolmogorov-Arnold Network (KAN) module, while the summarization part utilizes attention-based GRU, combined with a KAN model head. Classification model yields 67.96% of accuracy and 0.65 F1 score; while ROUGE-1, ROUGE-2, and ROUGE-L measures for summarization yield 0.38, 0.23, and 0.31 F1 scores, correspondingly. Ablation study shows that the use of KAN increases classification accuracy from 57.34% to 67.96%. Moreover, our proposed technique is compared to several baselines, including classical ML algorithms and pretrained language models.

Problem

• legal document classification
• text summarization
• low-resource multilingual
• class imbalance
• long dependencies

Innovation

• Kolmogorov-Arnold Network
• BiGRU
• legal document classification
• text summarization
• low-resource multilingual