ClaimGAT Graph Attention Network for Patent Litigation Prediction
August 25, 2026
ClaimGAT uses a directed dependency graph and GATConv layers to model relationships between independent and dependent patent claims. Testing on 1.34 million USPTO patents shows that per-claim encoding and graph connectivity improve litigation risk prediction accuracy.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates the predictive value of encoding legal dependency structures rather than using flat text sequences.