The construction method of traffic violation knowledge question answering system based on BERT-CRF model

In the context of AI+, a knowledge question and answer system for significant traffic offenses in Beijing was built utilizing deep learning technology, which can rapidly and effectively aid in handling traffic violations. First, the knowledge graph of traffic violation in Beijing was created using technologies such as knowledge acquisition, knowledge extraction, knowledge fusion, and knowledge storage. Second, the BERT-CRF (Bidirectional Encoder Representations from Transformers-Conditional Random Field) model was used to extract and parse natural language questions, while the knowledge graph was used for querying and reasoning to provide replies. To enable the knowledge question and answer system to accurately identify traffic violation incidents, use the BIO annotation method to expand the People’s Daily dataset and the BERT-CRF model was achieved a precision of 88.53%, a recall rate of 92.90%, and an F1 score of 90.66%. By leveraging the BERT-CRF model, the system can quickly and accurately return the queried answers to users. It has been verified that the application effect is good in terms of road traffic safety prompts.