BERT

BERT (Bidirectional Encoder Representations from Transformer) is a variant of the transformer architecture built upon the original encoder submodule. It specialises in masked word prediction and it is very strong in classification tasks, like sentiment prediction and document categorization.