App Relevance Classification on App Relevance Dataset (test)
0.9595PR-AUCFortress
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FortressTraining=Multi-snapshot, Features=Reduced feature set (unstable feature removal)2026.05 | 0.9595 | 0.5274 | |
| Multi-snapshot with all featuresTraining=Multi-snapshot, Features=SR + engagement2026.05 | 0.9593 | 0.5283 | |
| All features (SR + engagement)Features=SR + engagement2026.05 | 0.9572 | 0.5462 | |
| SR features (LLM + DistilBERT) onlyFeatures=Semantic Representation (LLM + DistilBERT)2026.05 | 0.9367 | 0.4744 |