Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Generalizable Heterogeneous Legal Link Prediction

Lorenz Wendlinger, Simon Alexander Nonn, Abdullah Al Zubaer, Michael Granitzer

TL;DR

This work addresses robust generalizable link prediction in large, heterogeneous legal citation graphs, including cross-jurisdictional data from New Zealand. It introduces Robust Heterogeneous Graph Enrichment (R-HGE), combining edge dropout, feature concatenation, and an asymmetric interleave decoder to improve robustness and predictive accuracy over prior methods. Extensive experiments on disjoint datasets OLD201k and LiO338k demonstrate substantial gains, with ablations showing edge dropout as a key contributor and transferability analyses revealing both opportunities and limits for inductive generalization across legal systems. The study extends evaluation to multilingual, cross-system contexts, offering practical insights for deploying robust legal link prediction in diverse, real-world settings.

Abstract

Recent work has applied link prediction to large heterogeneous legal citation networks \new{with rich meta-features}. We find that this approach can be improved by including edge dropout and feature concatenation for the learning of more robust representations, which reduces error rates by up to 45%. We also propose an approach based on multilingual node features with an improved asymmetric decoder for compatibility, which allows us to generalize and extend the prediction to more, geographically and linguistically disjoint, data from New Zealand. Our adaptations also improve inductive transferability between these disjoint legal systems.

Robust Generalizable Heterogeneous Legal Link Prediction

TL;DR

This work addresses robust generalizable link prediction in large, heterogeneous legal citation graphs, including cross-jurisdictional data from New Zealand. It introduces Robust Heterogeneous Graph Enrichment (R-HGE), combining edge dropout, feature concatenation, and an asymmetric interleave decoder to improve robustness and predictive accuracy over prior methods. Extensive experiments on disjoint datasets OLD201k and LiO338k demonstrate substantial gains, with ablations showing edge dropout as a key contributor and transferability analyses revealing both opportunities and limits for inductive generalization across legal systems. The study extends evaluation to multilingual, cross-system contexts, offering practical insights for deploying robust legal link prediction in diverse, real-world settings.

Abstract

Recent work has applied link prediction to large heterogeneous legal citation networks \new{with rich meta-features}. We find that this approach can be improved by including edge dropout and feature concatenation for the learning of more robust representations, which reduces error rates by up to 45%. We also propose an approach based on multilingual node features with an improved asymmetric decoder for compatibility, which allows us to generalize and extend the prediction to more, geographically and linguistically disjoint, data from New Zealand. Our adaptations also improve inductive transferability between these disjoint legal systems.
Paper Structure (8 sections, 3 figures, 4 tables)

This paper contains 8 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Robust Generalizable Legal Link Prediction
  3. Datasets
  4. Experiments
  5. Link Prediction
  6. Ablation Studies
  7. Transferability
  8. Conclusion

Figures (3)

  • Figure 1: R-HGE for a graph with some dropped edges (x), feature concatenation and asymmetric interleave decoder.
  • Figure 2: OLD201k and LiO338k Dataset Distributions.
  • Figure 3: Fully inductive transfer within and between datasets, colour indicates deviation from the semi-inductive score for on-diagonal elements and from the within-dataset score otherwise.