Text classification model for argument mining and detection

gbert-large-argument_mining is a text classification model in the scientific domain in German, finetuned from the model gbert-large. It was trained using a synthetically created, annotated dataset containing different sentence types occuring in conclusions of scientific theses and papers.

Training

Training was conducted on a 10 epoch fine-tuning approach, however this repository contains the results of the fourth epoch, since it has the best accuracy:

Epoch	Accuracy	Loss
1.0	0.9178	0.2491
2.0	0.9315	0.2479
3.0	0.9315	0.2853
4.0	0.9384	0.2503
5.0	0.9110	0.3678
6.0	0.9315	0.3436
7.0	0.9247	0.3807
8.0	0.9178	0.3862
9.0	0.9178	0.3953
10.0	0.9178	0.3964

In relation to the dataset, the model demonstrates that it can effectively learn to distinguish between the two classes claim and premise. However, the rapid onset of overfitting after epoch 4 suggests that the dataset is imbalanced and noisy. Further work should enable the model to be trained on more robust data to ensure better evaluation results.