microsoft/SportsBERT

SportsBERT is a BERT model trained from scratch with specific focus on sports articles . The training corpus included news articles scraped from the web related to sports from the past 4 years . There were approximately 8 million articles which were used to train this model . It's a MLM based transformers model and the primary task of the model is to fill in missing masked tokens . For example, "Anthony Davis is a [MASK]" would give out the tokens "legend", "superstar", "rookie", "star" in descending confidences . This model can then be used to fine tune for other tasks such as classification, entity extraction, etc. The model was trained,