Improved Named Entity Recognition in real-world applications for legal documents
Moreno-Acevedo SA, Orozco-Arroyave JR (2026)
Publication Type: Journal article
Publication year: 2026
Journal
Book Volume: 351
Article Number: 116877
DOI: 10.1016/j.knosys.2026.116877
Abstract
Named Entity Recognition (NER) is a topic of natural language processing (NLP) that has gained interest in the research community for its application in different law-related environments where the analysis of documents is crucial. So far, NER has been extensively studied in general contexts using non-realistic datasets, with many samples per class. However, typical real-world contexts do not have large amounts of data or balanced classes. This paper presents a methodology where these two problems are addressed. The approach includes a focal loss function incorporated into a transformer-based model and also the implementation of data augmentation techniques focused on NER data. The methodology is evaluated in three scenarios: (i) an ideal scenario with more than 11,000 samples and balanced classes with decisions of the German court; (ii) a standard scenario with a few samples, and classes with a medium level of imbalance from registrations of the Colombian Chamber of Commerce (CCC); and (iii) a scenario with a few samples from Spanish court decisions with a high imbalance among classes. The methodology yields promising and competitive results especially when data are scarce and highly imbalanced. In addition, we evaluated the generalization capability of the model from general to specific domains and from specific to specific domains. The model shows a promising knowledge generalization when considering different data domains, which is useful when working with reduced amounts of data.
Involved external institutions
Colombia (CO)How to cite
APA:
Moreno-Acevedo, S.A., & Orozco-Arroyave, J.R. (2026). Improved Named Entity Recognition in real-world applications for legal documents. Knowledge-Based Systems, 351. https://doi.org/10.1016/j.knosys.2026.116877
MLA:
Moreno-Acevedo, Santiago A., and Juan Rafael Orozco-Arroyave. "Improved Named Entity Recognition in real-world applications for legal documents." Knowledge-Based Systems 351 (2026).
BibTeX: Download