An mDeBERTa-v3 model trained to recognize whether a paragraph contains coverage of conflict between political actors. Fine-Tuned on 9500 English and Spanish paragraphs taken from media sources in the US, UK, Canada, Australia, Mexico and Spain.
Performance:
Validation Accuracy: .894 Validation F1 Macro: .805 Valdiation Balanced Accuracy: .792
Test set Accuracy = .887 On a Bilingual test set, the model performs equally well in English and Spanish (English Accuracy = .891, Balanced Accuracy = .781, Spanish Accuracy = .880, Balanced Accuracy = .813)