| In recent years,due to the rapid development of the Internet industry,the integration of the Internet industry and the financial industry has been promoted.As a result,thousands of Internet financial enterprises have come into being,bringing with them a huge amount of natural language data related to Internet finance.Chinese named entity recognition for massive Internet financial data has become an indispensable part of the field of natural language processing.Named entity recognition is to point to in the text in the text,to identify the specific entity,it is one of the most basic problem in natural language processing field,now known as the named entity recognition model of Internet financial entity is not efficient identification,if not accurate identification of the Internet financial entities,will affect the national supervision of financial enterprises on the Internet.In recent years,entity recognition schemes based on deep learning have gradually become outstanding in the field of traditional named entity recognition.However,there are still some problems in the task of Chinese named entity recognition in specific fields.Traditional models cannot effectively solve the problem that a word in a Chinese sentence contains multiple meanings.Specific knowledge is required to identify specific domains such as Internet finance.Based on the above problems,this thesis improves the traditional named entity model based on deep learning,uses professional Internet financial data sets for training and recognition,and finally constructs a set of Internet financial entity recognition system based on deep learning.The specific work is as follows:Firstly,in view of the lack of a method to obtain Chinese word semantic information in the benchmark model Bert-Bi LSTM-CRF,the Bert-WWM model based on the full-word coverage strategy is introduced to effectively obtain Chinese word-level semantic information.Second,aiming at the problem that the traditional Bi LSTM model takes too long to calculate due to too many parameters and too complex model,this thesis introduces Bi GRU model to reduce the model parameters,which can greatly reduce the calculation amount of the model.Thirdly,this thesis proposes an Internet financial entity identification model based on Bert-WWM-Bi GRU-Att-CRF.In order to solve the problem of distraction in the coding layer of Bi GRU network for the current input sequence,an attention mechanism layer is added to the existing model,which can effectively screen out the important information and the sub-important information in the same input sequence.Fourthly,through the entity recognition model proposed above,this thesis builds the Internet financial entity recognition system,which can first crawl the huge financial information on the network through the network crawler,and then carry out the Internet financial entity recognition through the above entity recognition model.The experimental results show that the proposed model has more advantages than the existing models,and the F1 value is increased by 1.99% compared with the baseline. |