The BERT-BiLSTM-CRF Question Event Information Extraction Method

The purpose of this research is to further improve the accuracy of the entity information extraction of customer voice consultation questions, enhance the overall effect of information extraction technology in the knowledge map of the intelligent question and answer system, optimize information from the aspect of semantic labelling, and add the BERT model on the basis of BiLSTM-CRF to study sentences in depth, thus enhancing the semantic information extraction of questions. Entity extraction and semantic labelling experiments are carried out with a data source of customer's voice consultation questions and question text. The results show that the semantic label extraction tuning average of the BERT model is higher than that of the BiLSTM-CRF method on the basis of semantic label optimization, and the model tuning average is 91.53%, which can provide practical significance for enhancing event entity information extraction.