terran-ai/WOS46985Bert

BERT classifier for WOS-46985

This is a model to classify scientific papers by the Web-of-Science nomenclature.

Model Details

Model Description

It's a fine-tuned model to predict the 134 classes from the WOS-46985 model published by https://arxiv.org/pdf/1709.08267.pdf.

• Developed by: Terran (https://terran.ai/) SciLifeLab Data Center (https://www.scilifelab.se/) and KTH Research Support Office (https://intra.kth.se/en/styrning/kths-organisation/vs/rso).
• License: Apache 2.0
• Finetuned from model: bert-base-uncased

Evaluation

10/90 validation/training split (like https://arxiv.org/pdf/1709.08267.pdf)

Results

Accuracy on the final layer was 83% (previous state-of-the-art 77% https://arxiv.org/pdf/1709.08267.pdf). However, the previous SOTA did not use test-data set, so the difference is probably more significant.

Summary

Useful model to annotate scientific text.