Data Science2024
Fine-Tuning BERT for Domain-Specific Named Entity Recognition
Adapted a pre-trained BERT model for Named Entity Recognition on Lord of the Rings text, identifying characters, locations, and organizations.
Methodology
Comparative training on the English Web Treebank versus Lord of the Rings-specific data to assess the effect of domain adaptation.
Findings
Domain-specific training data produced significant performance improvements over general-purpose data alone.
Tools & Methods
BERTPyTorchTokenizationF1 / Precision / Recall