NYU NLP Final Project: Build a semantic role labeling system utilizing SOTA machine learning models
- Ziyang Zeng: Maxent Baseline, DistilBERT related experiments
- Jiahao Chen: Random Forest
- Peiwen Tang: RoBERTA, Feature engineering, Word2Vec and downstream experiments
- Zeyu Yang: BERT base model
- Baseline: Maxent
- Word2Vec: Feature Extraction and Classification
- Random Forest: Notebook
- BERT (base): Notebook
- DistilBERT: Notebook
- RoBERTA: Notebook
- DistilBERT (POS+BIO): Notebook
- DistilBERT (QA): Notebook
- DistilBERT (ONE ARG1): Notebook
On %-test:
| Model | Precision | Recall | F1 | Output |
|---|---|---|---|---|
| Maxent | 71.88 | 61.33 | 66.19 | txt |
| RandomForest | 64.53 | 74.00 | 68.94 | txt |
| BERT | 91.33 | 91.33 | 91.33 | txt |
| DistilBERT | 93.75 | 90.00 | 91.84 | txt |
| RoBERTA | 91.50 | 93.33 | 92.41 | txt |
| DistilBERT (POS+BIO) | 93.19 | 91.33 | 92.25 | txt |
| DistilBERT (QA) | 92.00 | 92.00 | 92.00 | txt |
| DistilBERT (ONE ARG1) | 92.67 | 92.67 | 92.67 | txt |
On total-test:
| Model | Precision | Recall | F1 | Output |
|---|---|---|---|---|
| Maxent | 55.33 | 36.02 | 43.64 | txt |
| DistilBERT | 80.49 | 78.43 | 79.45 | txt |