An enterprise entity relation extraction model based on ERNIE

In the field of enterprise and economics, there are a lot of unstructured text data of investment, industry and commerce. These data often exist in the form of triples of “entity”, “object” and “relationship”. For the information and data extraction of such problems, this paper proposes a financial entity relationship extraction model based on ERNIE. ERNIE was used for pre-training, BIO mode was used for annotation, and coding sequence was obtained. The triplet containing subject, object and predicate is extracted by the semantic representation of context, and the subject, object and relation are extracted by sigmoid in the prediction layer, and the coding sequence is decoded to obtain the final processing result. Experimental results show that in the data set of manually annotated investment events and industrial and commercial events, the effect of this model is significantly better than the traditional BERT and CNN-Attention based joint entity relation extraction model, and the F1 value used to evaluate the model performance reaches 81.72%.