The current model is trained using non-empty pages on which both annotators agreed.
Include pages with annotator disagreement into the training data (with weighted and/or average score) could improve the results.
[ ] Adapt training data selection method
[ ] Retrain model
[ ] Evaluate new model, and compare to current results
The current model is trained using non-empty pages on which both annotators agreed. Include pages with annotator disagreement into the training data (with weighted and/or average score) could improve the results.