nyu-nlp-final-project

NYU NLP Final Project: Build a semantic role labeling system utilizing SOTA machine learning models

Authors

• 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

Experiments

• 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

Results

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