3-class classification on Narratives 54 narratives, 48,640 samples (held-out)
57.8Aligned AccuracySVAF (neural)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SVAF (neural)learned_encoder=true, cross-field_attention=true, fusion_gate=neural, cross-domain_relevance_prior=true2026.04 | 57.8 | 87.7 | 70.5 | 78.7 | |
| Heuristic per-field (αf)fusion_gate=weighted cosine averaging2026.04 | 48.9 | 83.4 | 63.2 | 73.1 | |
| Scalar + temporal decaytemporal_decay=true2026.04 | 43.5 | 80.1 | 55.8 | 68.4 | |
| Scalartype=cosine, theta (θ)=0.52026.04 | 41.2 | 79.3 | 52.1 | 66.8 |