Topic Shift Detection on Topic Shift Dataset 300 < ΔT ≤ 512 v1.0 (test)
69.7PrecisionNonlinear Naive Bayes With Attention
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Nonlinear Naive Bayes With AttentionMechanism=Attention-based NLU, Token Gap Range=300 < ΔT ≤ 5122026.02 | 69.7 | 97.2 | 77.5 | 81.2 | |
| NSPModel Type=BERT-based NSP, Token Gap Range=300 < ΔT ≤ 5122026.02 | 61.2 | 98.2 | 67.9 | 75.4 |