A systematical procedure to extracting legal entities from Indonesian judicial decisions

TL;DR

The paper introduces a systematic method for extracting legal entities from Indonesian court decisions using named entity recognition, aiming to improve legal transparency and analysis.

Contribution

A structured pipeline for gathering, filtering, and annotating Indonesian judicial decisions with over 50 legal entity types using NER techniques.

Findings

A dataset of 2687 annotated court documents was created using manual annotation and inter-annotator agreement (Fleiss' Kappa score of 0.705).

The method supports downstream NLP tasks like legal information extraction and classification through token-level translation and BIO tagging.

Abstract

This article presents a systematic method of extracting legal entities from Indonesian judicial decisions with a well-structured named entity recognition (NER) approach. The procedure was implemented by gathering and annotating court decisions for theft cases at three court levels: first instance (2478 files), appeal (147 files), and cassation (62 files), amounting to 2687 annotated files. The data were harvested from the official website of the Supreme Court of the Republic of Indonesia using automated web scraping, followed by manual filtering for relevance and completeness.

Manual annotation was performed with the Label Studio platform by three independent annotators. Annotation consistency was considered using Fleiss' Kappa, yielding an average agreement score of 0.705 across all levels, indicating good inter-annotator reliability. The method uses a hierarchical structure and a BIO tagging scheme to tag >50 types of legal entities, including defendants, judges, legal articles, charges, and verdict decisions.

This approach is proper for text processes such as legal information extraction, classification, and legal analysis. From a legal perspective, this process will improve legal transparency and research on Indonesian judicial data.•Structured pipeline for gathering, filtering, and annotating Indonesian court judgments based on legal metadata and web scraping.•Manual annotation of 2687 court documents with annotation rules and inter-annotator agreement using Fleiss' Kappa.•Token-level translation and BIO tagging for >50 legal entities, enabling downstream NLP tasks such as named entity recognition.

Structured pipeline for gathering, filtering, and annotating Indonesian court judgments based on legal metadata and web scraping.

Manual annotation of 2687 court documents with annotation rules and inter-annotator agreement using Fleiss' Kappa.

Token-level translation and BIO tagging for >50 legal entities, enabling downstream NLP tasks such as named entity recognition.

Image, graphical abstract

Figures