PCSciBERT_cased was initiated with the cased variant of SciBERT (https://huggingface.co/allenai/scibert_scivocab_cased) and pre-trained on texts from 1,560,661 research articles of the physics and computer science domain in arXiv. The tokenizer for PCSciBERT_cased uses the same vocabulary from allenai/scibert_scivocab_cased.
The model was also evaluated on its downstream performance in named entity recognition using the adsabs/WIESP2022-NER and CS-NER (https://github.com/jd-coderepos/contributions-ner-cs/tree/main) dataset. Overall, PCSciBERT_cased achieved higher micro F1 scores than SciBERT(cased) for both WIESP (Micro F1: 82.19%) and CS-NER (Micro F1: 76.22%) datasets.
It improves the performance of SciBERT(cased) on CS-NER test dataset by 0.69% and on WIESP test dataset by 1.49%.
- Downloads last month
- 5