Research On Named Entity Recognition Method In Civil Aviation Business

On the premise of ensuring safety,improving passenger service quality is an important support point for increasing transportation volume.Obtaining accurate identification of the pain points in the service link has become a prerequisite for improving the quality of passenger service.Passengers are used to evaluate the service on the Internet.Airport passenger service evaluation is our research object.We study the passenger evaluation texts related to the airport service through the platforms of post bar,blog and comment website.These evaluation texts include all kinds of services enjoyed by passengers at the airport.Civil aviation passenger evaluation text has many sources and poor standardization.The existing named entity recognition technology can not find the passenger evaluation object from these texts very well.This thesis explores the application and improvement of entity recognition technology in civil aviation.Aiming at the lack of standard data sets in the civil aviation,this thesis uses a combination of manual labeling and semi-supervised learning to construct a rare standard data set.Aiming at the characteristics of a large number of composite entities in the passenger evaluation text,this thesis enhances the model’s ability to extract local information on the basis of the BLSTM and a model incorporating CNN is proposed.The local features extracted by ODCNN are fused with the long-dependent features extracted by BLSTM.Then an ECNN recognition method is proposed,which uses convolution kernels of different sizes to extract local information in different ranges to enhance the recognition performance of the model again.Aiming at the imbalanced characteristics of passenger evaluation data,a multi-agent fusion neural network entity recognition framework is designed,which uses knowledge resources and uses different neural networks to recognize the entities that they are good at.The experimental results show that the method proposed in this thesis surpasses the basic BLSTM-CRF model in terms of F value,and can better recognize airport business entities.