Research on entity recognition technology and application for power customer service work orders

This paper proposes an entity recognition method for power customer service work orders based on the BERTBiLSTM -CRF model. The experimental data set contains 45,900 sentences, divided into 8 types of work orders, annotated with special entities and 3 common entities (account number, telephone number, address). BIOSE annotation is used, and precision, recall and F1 value are used as evaluation indicators. The results show that the model has a good recognition effect on common entities such as account number, telephone number and address, with F1 values of ${9 2. 7 8 \%}, {8 9. 5 0 \%}$ and ${8 8. 0 1 \%}$ respectively; the recognition effect is better for work orders with large data volume and complete structure, which provides support for the power system to quickly extract abnormal power consumption information of customers.