SoftLexicon-BERT-BiLSTM-CRF-based Named Entity Recognition Model for Transportation Organization Design Domain

The traffic organization design field involves more disciplines and the same entity is expressed in a variety of ways, which causes the named entity recognition effect to be greatly reduced, and the existing named entity recognition method based on single character granularity is unable to solve the dissimilarity of characters in different traffic organization vocabularies. In order to solve this problem, an entity extraction method based on the enhancement of domain knowledge features is investigated, which incorporates the domain knowledge system and professional vocabularies as external information into the domain entity recognition process of intersection traffic organization design, so as to improve the accuracy of named entity recognition. A traffic organization design domain entity extraction model based on the SoftLexicon-BERT-BiLSTM-CRF framework is constructed, and experiments on textual named entity extraction in traffic organization design domain are carried out, and the results show that, relative to the Bi LSTM-CRF model, the accuracy of this paper's model is improved by about 7.5%, the recall rate is improved by about 11%, and the F1-score improved by about 9.4%; relative to the BERT-BiLSTM-CRF model, the model in this paper improves the accuracy by about 3%, the recall by about 4%, and the F1-score by about 4%.