TriView-GNN: Legal-Factor Graph Modeling for Tripartite Viewpoint Classification in Labor Dispute Judgments

杨阳

Keywords: Labor Dispute; Tripartite Viewpoint Classification; Large Language Models (LLMs); Graph Neural Networks (GNNs); Discourse Logic; Legal Factors

Paper Type: Short papers / work-in-progress

TL;DR: We propose TriView-GNN to accurately classify employee, employer, and court viewpoints in labor dispute judgments.

Abstract:

Focusing on the high incidence of labor dispute cases in contemporary judicial practice, this paper investigates the core challenge faced by Large Language Models (LLMs) in processing the corresponding adjudicative documents: the precise classification of the tripartite viewpoints of the "employee, employer, and court." Due to their unique tripartite structure, ambiguous and flexible legal terminology, and complex argumentative logic, labor dispute documents pose a formidable test for current natural language processing technologies. Existing models face three primary bottlenecks: (1) Domain Mismatch in Foundational Parsing, where general-purpose word segmentation tools fail to accurately process compound legal terms; (2) Failure to Model Discourse Logic and Legal Reasoning, with models struggling to capture the long-range "claim-defense-finding" argumentative chain; and (3) Severe Scarcity of High-Quality Annotated Data, where prohibitive costs and annotation complexity constrain model training.
To address these challenges, this paper proposes a systematic set of countermeasures. First, we advocate for a specialized text parsing system for the labor law domain to improve the accuracy of terminological recognition. Second, we innovatively introduce Graph Neural Networks (GNNs) to deconstruct adjudicative documents into an argumentation graph composed of "legal factors," thereby explicitly modeling discourse logic and the judicial syllogism. Finally, we propose a data strategy that combines semi-supervised and active learning to efficiently construct a large-scale, high-quality training corpus while controlling costs. This research aims to provide a viable technical path for LLMs to transition from general language understanding to specialized judicial reasoning, holding significant theoretical and practical importance for enhancing the precision and interpretability of judicial intelligence in the field of labor disputes.

Submission Number: 5