An End to End joint entity recognition and relationship extraction based on multi-domain Chinese corpus

Chinese sentences are flexible and complex, and there are few structured data in Chinese text, which brings great challenges to Information Extraction. Besides, most of the current methods of NER and RE are concentrated in the medical field. In the later part, we present a new joint entity and relationship extraction method. First of all, we use ALBERT to obtain the embedded vector of each character, input it into the multilayer bidirectional transformer encoder to get representation of the text vector. Then use BilSTM to calculate the hidden state of each character in Chinese sentence, and use Dense to calculate the hidden vector of the sentence. Finally, CRF is used to extract the relationship. At the same time, we conducted the experiment, we calculated the F1 score and recall rate of each relation, and both macro avg and weighted avg scores were improved to a certain extent.