FinBERT–MRC: Financial Named Entity Recognition Using BERT Under the Machine Reading Comprehension Paradigm

Financial named entity recognition (FinNER) is a challenging task in the field of financial text information extraction, which aims to extract large amount knowledge from unstructured texts. It widely accepted use sequence tagging framework implement FinNER tasks. However, such models cannot fully take advantage semantic Instead, we formulate as machine reading comprehension (MRC) problem and propose new model termed FinBERT–MRC. This formulation introduces significant prior by utilizing well-designed queries, extracts start index end target entities without decoding modules conditional random fields (CRFs). We conduct experiments on publicly available Chinese dataset ChFinAnn real-world business AdminPunish. FinBERT–MRC achieves average $${F}_{1}$$ scores 92.78% 96.80% two datasets, respectively, with gains + 3.75% 0.68% over some including BiLSTM–CRF, BiGRU–CRF, BiLSTM–CNN–CRF, FinBERT–Tagger, FinBERT–CRF. The source code at https://github.com/zyz0000/FinBERT-MRC .