ConfliBERT is a pre-trained language model for political conflict and violence. We provided four versions of ConfliBERT: <ol> <li>ConfliBERT-scr-uncased: Pretraining from scratch with our own uncased vocabulary (preferred)</li> <li>ConfliBERT-scr-cased: Pretraining from scratch with our own cased vocabulary</li> <li>ConfliBERT-cont-uncased: Continual pretraining with original BERT's uncased vocabulary</li> <li>ConfliBERT-cont-cased: Continual pretraining with original BERT's cased vocabulary</li> </ol>
See more details in https://github.com/eventdata/ConfliBERT/