Entity relationship extraction model based on RoBERTa and pointer annotation network

In order to solve the problems of insufficient sentence feature extraction in traditional models and poor recognition of overlapping entities by previous models, this paper proposes to use the RoBERTa pre-training model with stronger sentence feature extraction capabilities and the pointer annotation network with stronger overlapping entity recognition capabilities for entity recognition. Relation extraction. This method uses the RoBERTa pre-training model to extract richer sentence features, and uses pointer annotation to mark the starting position and ending position of entities on the same two sequences, thereby effectively extracting overlapping entities. Through experimental verification, the model in this paper has better performance improvement on the DuIE benchmark data set.