List your company

Hallucination Detection on CWQ (F1, Precision, Recall, Accuracy, AUC-PR)

87.4F1 ScoreKG-Guard
57.13664.99372.8580.707May 29, 2026

Evaluation results

MethodLinks
KG-GuardModel Backbone=GraphTransformer2026.0587.485.890.283.594.2
KG-GuardModel Backbone=GAT2026.0586.384.589.681.793.1
GPT 5 MiniInput Context=G+RS+RT2026.0584.293.576.779.488.4
Most Frequent2026.0583.571.710071.771.7
GPT 5 MiniInput Context=G2026.0582.690.276.27785.8
SelfCheckSampling size (N)=102026.0572.290.460.266.983
SelfCheckSampling size (N)=52026.0567.19153.162.681.9
Llama 4 ScoutInput Context=G+RS2026.0561.39445.458.881.8
Llama 4 ScoutInput Context=G+RS+RT2026.0561.294.445.358.882
Llama 4 ScoutInput Context=G2026.0560.291.644.857.580.7
Random2026.0558.371.449.349.471.6