Introducing KAN Block with BiGRU for Legal Document Classification and Summarization

ACL ARR 2026 January Submission6735 Authors

Keywords: Legal Document Classification, Legal Text Summarization, BiGRU, KAN Block, Class Imbalance, Deep Learning

Abstract: This work introduces a new deep learning based solution for legal document classification and summarization (LDCS), which is capable of tackling several challenging issues, including domain-specific language, long-term dependencies and class imbalance. To make model better process complex legal texts, a Kolmogorov–Arnold Network (KAN) block is added to the model. Extensive experiments on a legal dataset show the effectiveness of the proposed approach compared to standard ML models (SVM and logistic regression), with 67% classification accuracy, weighted F1-score of 0.65, and summarization ROUGE-1 F1-score = 0.38. These findings demonstrate the promise of the approach for further developing automated legal analysis and decision support systems.

Paper Type: Long

Research Area: Interpretability and Analysis of Models for NLP

Research Area Keywords: Information Extraction and Retrieval, Summarization, Machine Learning for NLP, Interpretability and Analysis of Models for NLP, Natural Language Generation, Ethics, Bias, and Fairness

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 6735