JudgmentGraph: Benchmarking Legal Knowledge Graph Construction from Supreme Court Rulings

Legal judgments are rich in factual, procedural, and normative content, yet are typically presented in unstructured natural language, limiting their accessibility for automated reasoning and retrieval. This paper proposes a comprehensive and ontology-guided pipeline for constructing a Legal Knowledge Graph from Indian Supreme Court judgments. The pipeline integrates a fine-grained Legal Named Entity Recognition system with a formally defined Legal Ontology to generate semantically valid triples. A multi-stage relation extraction module leverages syntactic cues, proximity heuristics, and logical constraints to refine relation candidates. The resulting triples are serialized into RDF and visualized using WebVOWL. Empirical evaluation on a manually annotated gold-standard corpus demonstrates strong performance with a Precision of 81.4%, Recall of 88.4%, and an F1-score of 84.7%. A qualitative case study on the Lalita Kumari v. Government of Uttar Pradesh ruling illustrates the interpretability and domain-alignment of the extracted graph. This work contributes the first benchmarked pipeline for legal knowledge graph construction in the Indian legal domain and lays the foundation for downstream applications in legal information retrieval, reasoning, and decision support.