CS334-Final-Project

Datasets

Extracted from WSD Evaluation Framework: http://lcl.uniroma1.it/wsdeval/

Train: Semcor

Validation: SE7

Test: SE2, SE3, SE13, SE15

Model structure

Words with multiple senses are called polysemes.

Sentences -> Sentence Embeddings (WordPiece) -> Encoder (BERT transformer) -> Features (vectors) of polysemes in sentence

Features -> Classifier (MLP neural net) -> Predictions of senses

Training the Model

Look at sections 3.1 and 3.2 in the WSD Using BERT paper