PRecG: Legal Precedent Retrieval with Graph Neural Networks and Rhetorical Role Segmentation

Short summary

PRecG is a pipeline for legal precedent retrieval that decomposes judgments into rhetorical-role segments, builds knowledge graphs for each segment to capture legal entities and relationships, and hierarchically aggregates segment embeddings into document-level representations for similarity computation. The approach addresses the limitation of treating legal documents as monolithic texts, which overlooks contextual significance of legal concepts. Experiments on an Indian legal benchmark dataset demonstrate effectiveness over state-of-the-art baselines.

• •Decomposes legal documents into rhetorical-role segments for finer-grained representation
• •Constructs per-segment knowledge graphs capturing legal entities and relationships
• •Validated on Indian legal dataset, outperforming existing similarity-based retrieval methods

Generated with AI, which can make mistakes.