Legal judgment prediction via text classification
Journal of King Saud University Computer and Information Sciences · 2026-11 · 38 (9)
DOI: 10.1007/s44443-026-01360-8 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of King Saud University Computer and Information Sciences, online erschienen am 04.11.2026. Verfasst von Yabo Liu und 1 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Natural Science Foundation of Hebei Province, Science and Technology Cooperation Special Project of Shijiazhuang.
Abstract (Originalsprache)
Legal judgment prediction (LJP) is a key task for assisting judicial decision making, aiming to predict charges, relevant legal articles, and terms of penalty from factual descriptions of cases. Although Fast Linear Attention with a Single Head (FLASH) enables efficient sequence modeling with linear complexity, its representation capability is limited in capturing the semantics of legal texts. To address this issue, we propose EMAgent-FLASH, an enhanced FLASH architecture that incorporates bidirectional Exponential Moving Average (EMA) layers and a Depthwise Convolution (DWC) module. The EMA layers generate dynamic agent tokens to facilitate information interaction between local quadratic and global linear attention, while the DWC module enhances local contextual feature extraction. Based on EMAgent-FLASH, we develop EMAgent-FLASH-CM for LJP, which adopts a hybrid pooling strategy to combine CLS representations with attention-based features and employs a symmetric loss function to improve classification boundary learning under class imbalance. Experimental results on the CAIL2018-Small dataset show that EMAgent-FLASH-CM achieves Macro-F1 scores of 84.99% and 84.69% for law article and charge prediction, respectively. Further evaluations on AG News and Reuters-21578 show that the model achieves Macro-F1 and Micro-F1 scores of 96.89% and 91.15%, respectively, for multi-class and multi-label classification. These results demonstrate that the proposed model enhances text representation and exhibits applicability across different classification scenarios.
Förderung
- Natural Science Foundation of Hebei Province
- Science and Technology Cooperation Special Project of Shijiazhuang
Zitieren
APA
Liu, Y., & He, H. (2026). Legal judgment prediction via text classification. Journal of King Saud University Computer and Information Sciences, 38(9). https://doi.org/10.1007/s44443-026-01360-8
Vancouver
Liu Y, He H. Legal judgment prediction via text classification. Journal of King Saud University Computer and Information Sciences. 2026;38(9). doi:10.1007/s44443-026-01360-8
BibTeX
@article{Liu2026,
author = {Liu, Yabo and He, Hao},
title = {Legal judgment prediction via text classification},
journal = {Journal of King Saud University Computer and Information Sciences},
year = {2026},
volume = {38},
number = {9},
doi = {10.1007/s44443-026-01360-8}
}
EUVIMED