TLDR: The paper introduces a novel Legal Knowledge Graph (KG) focused on legislation concerning violence against women. It details two automated methodologies for KG construction: a precise, systematic bottom-up approach and a scalable, flexible Large Language Model (LLM)-based approach. The resulting KG, derived from European Court of Human Rights judgments, aims to improve access to legal information, enable complex queries, and support predictive justice applications, adhering to FAIR data principles.
In the complex world of legal decision-making, having access to comprehensive and up-to-date legislative information is crucial. However, structured and easily queryable resources, like Legal Knowledge Graphs (KGs), are often scarce in the legal domain. This gap is particularly significant when dealing with sensitive and widespread issues such as violence against women, which has profound impacts globally.
A recent research paper, titled “Automated Creation of the Legal Knowledge Graph Addressing Legislation on Violence Against Women: Resource, Methodology and Lessons Learned,” introduces a groundbreaking Legal Knowledge Graph specifically designed to address legislation and judicial cases related to violence against women. This innovative resource, along with its development methodologies, aims to enhance access to legal information for both humans and machines, facilitate complex queries, and serve as a vital component for machine learning tools in predictive justice.
Two Paths to Knowledge: Bottom-Up and LLM-Based Approaches
The researchers explored two distinct, yet complementary, automated methodologies for constructing this Legal KG:
The first is a systematic bottom-up approach, which is a customization of a general KG development process tailored for the legal domain. This method involves a structured pipeline starting with data collection from publicly available legal sentences, specifically 73 judgments and decisions from the European Court of Human Rights (ECHR). Experts in international law carefully selected these documents. Following data collection, knowledge is extracted from these legal texts, processed, and converted into “triples” – fundamental units of information in a KG. An ontology, which defines the concepts and relationships within the legal domain, is then created, and the KG is constructed by combining these triples and linking them to external resources like Wikidata. This approach is known for its high precision and semantic alignment, making it reliable for domain-specific and structured tasks. It resulted in over 10,000 extracted triples, representing a rich dataset for legal analysis.
The second methodology leverages the power of Large Language Models (LLMs), such as GPT-4.o and Mixtral. This approach is designed for automated KG generation, taking advantage of recent advancements in AI to streamline the process. It involves preparing legal documents, developing Retrieval-Augmented Generation (RAG) models to provide LLMs with specific context, creating a foundational ontology using LLMs, and then generating the KG. This method is highly scalable and adaptable, capable of processing large volumes of legal texts quickly. While it offers speed and flexibility, the LLM-based approach can sometimes face challenges with accuracy and consistency, requiring careful validation to prevent “hallucinations” or incomplete information. The paper evaluated this approach by answering competency questions, finding that processing the full text provided broader coverage, while focusing on specific sub-parts yielded more targeted results.
Complementary Strengths for a Robust Resource
The study highlights that both methodologies offer unique advantages. The bottom-up approach excels in precision and consistency, making it ideal for formal legal reasoning and detailed queries. In contrast, the LLM-based approach shines in its scalability and speed, making it suitable for exploratory tasks and rapid prototyping. The combination of these methods suggests a promising future for semi-automated legal knowledge engineering, where the strengths of structured extraction and advanced AI can be harmonized.
Also Read:
- The Fair Game: A Dynamic Approach to Ensuring AI Fairness Over Time
- New Insights into Multi-Winner Voting Through Data Analysis
Ensuring Accessibility and Reusability: The FAIR Principles
A key aspect of this Legal KG is its adherence to the FAIR data principles: Findable, Accessible, Interoperable, and Reusable. The KG is assigned a persistent DOI and registered in the LOD Cloud, ensuring it can be easily found. It is openly accessible via platforms like Zenodo and GitHub, and exposed through a public SPARQL endpoint for both human and machine access. Published in standard RDF/Turtle format and relying on widely adopted vocabularies, it ensures interoperability with other legal datasets. Finally, released under a Creative Commons Attribution 4.0 International License with full documentation and source code, it is designed for broad reuse and extension in legal AI applications.
This pioneering work represents a significant step forward in making legal information more structured, accessible, and actionable, particularly for critical issues like violence against women. The researchers plan to merge the two KGs into a unified resource and explore its applicability to other legal domains, contributing to a more unified and efficient representation of European laws. You can find the full research paper here.
