Rumor Detection on Arabizi dataset 1.0 (test)
84AccuracyDziriBERT FT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DziriBERT FTFamily=Transformers, Evaluation protocol=fine-tuned, Epochs=10, Batch size=32, Learning rate=enc: 1e-5 / clf: 5e-5, max len=1282026.06 | 84 | 85 | 84 | 84 | |
| DziriBERT + LRFamily=Hybrid, C=1, solver=lbfgs2026.06 | 84 | 84 | 84 | 84 | |
| DziriBERT + RFFamily=Hybrid, n_estimators=2002026.06 | 84 | 85 | 84 | 84 | |
| DziriBERT + SVMFamily=Hybrid, C=0.01, kernel=linear2026.06 | 83 | 83 | 83 | 83 | |
| mBERTFamily=Transformers, Epochs=8, Batch size=32, Learning rate=enc: 1e-5 / clf: 5e-5, max len=1282026.06 | 80 | 81 | 80 | 80 | |
| BiGRU + FastTextFamily=Deep Learning, Feature representation=FastText, Epochs=20, Batch size=32, Learning rate=1e-3, dim=128, patience=42026.06 | 79 | 79 | 79 | 79 | |
| SVM + TF-IDFFamily=Classical ML, Feature representation=TF-IDF, C=12026.06 | 77 | 77 | 77 | 77 | |
| LR + TF-IDFFamily=Classical ML, Feature representation=TF-IDF, C=10, solver=lbfgs2026.06 | 76 | 76 | 76 | 76 | |
| RF + TF-IDFFamily=Classical ML, Feature representation=TF-IDF, n_estimators=3002026.06 | 76 | 76 | 76 | 76 | |
| BiLSTM + FastTextFamily=Deep Learning, Feature representation=FastText, Epochs=20, Batch size=32, Learning rate=1e-3, dim=128, patience=42026.06 | 76 | 76 | 75 | 75 | |
| Llama3 (fine-tuned)Family=LLM, Evaluation protocol=fine-tuned, Epochs=5, Batch size=8, Learning rate=1e-4, grad. accum.=4, precision=bf162026.06 | 72 | 72 | 72 | 72 | |
| Llama3 (zero-shot)Family=LLM, Evaluation protocol=zero-shot, temp=0, API=Groq API2026.06 | 51 | 56 | 52 | 43 |