ERNIE-based Named Entity Recognition Method for Traffic Accident Cases

Abstract Traffic accident case named entity recognition, which helps mine key information in traffic texts, plays a vital role downstream tasks such as the construction of knowledge graphs road and intelligent policing. In this paper, we construct recognition model based on EDE (Entity Data Enhancement)-ERNIE-Bidirectional Gated Recurrent Unit Network (BiGRU)-Conditional Random Field (CRF) to address current situation low data poor long-text entities. First, amount is enhanced using random substitution method. Next, text cases are characterized dynamic word vector ERNIE pretraining model. Then, BiGRU network learns long-distance dependency relationship enhance effect recognition. Finally, result sequence constrained by CRF layer realize The experimental part uses related real domestic area. enhancement method increases volume three times compared original volume. Experimental results show that EDE-ERNIE-BiGRU-CRF achieves better F1 values, recall precision achieved performance than methods BERT-BiGRU-CRF, ERNIE-BiGRU-CRF, ERNIE-BiLSTM-CRF, ERNIE-CRF, ERNIE, BiGRU-CRF, ROBERTA-wwm-ext-BiGRU-CRF verify its effectiveness for cases.