From BERT to T5: A Study of Named Entity Recognition

In three linesComparative study of BERT (encoder-only) and T5 (seq2seq) for named entity recognition (NER). BERT uses a classification head with weighted cross-entropy; T5 fine-tuned with few-shot prompts. Evaluation on 7-class and simplified 3-class tag schemes, including ablation study and error analysis.

Summary generated by Claude — human-verified