dvdblk/scibert_sdg_cased_zo-up

SciBERT fine-tuned on SDG data

SciBERT fine-tuned on a custom dataset to predict the correct SDG class based on a scientific publication abstract.

Data

The training set consisted of 630 pairs of (abstract, sdg) with a balanced class distribution. The test set contains 271 samples.

Performance

Metric	Value
Accuracy	0.7528
F1	0.7465
Recall	0.7528
Precision	0.7507