Research On The Application Of Deep Learning Models In Geographic Named Entity Recognition

In the current information age environment,the scale of data as an information carrier continues to expand.In the massive text data description,there are a large number of geographically named entity names.With the development of natural language processing technology,the technology for obtaining domain target information from text data has gradually matured.Natural language is an important manifestation of the results of human cognition of the geographical world.How to obtain geographic entity information from unstructured text data is one of the difficult problems encountered by geographic information science.Geographically named entities are generally the center of the subject of description expression in natural language descriptions,just like the subject of a sentence.Therefore,obtaining the name of the geographic entity in the text is the key to obtaining the corresponding geographic information.With the mature application of Chinese named entity recognition technology,the need to realize the task of geographically named entity recognition is gradually being proposed.Geographically named entity recognition technology is the basis for obtaining geographic entity information from text data.This article is based on the widely used natural language processing model BERT(Bidirectional Encoder Representation from Transformers)model,combined with the long short-term memory model(Long Short-Term Memory,LSTM)and the random vector field model(Conditional Random Fields,CRF).Features,realizes the recognition of geographically named entity names from unstructured text data.The main research contents include:(1)First,according to the principle of Chinese named entity recognition,use the identification method based on rule matching and statistics to conduct named entity recognition research on experimental data.Analyze the description and expression characteristics of geographically named entities in Chinese,design the transformation method of text representation according to the characteristics of geographical entity names,and summarize the steps of geographically named entity recognition tasks and the problems that need to be solved,in order to use deep learning methods for geographically named entities recognition Provide basis for selection of algorithm principle.(2)The traditional name entity recognition method relies on the characteristics of the external knowledge base and the problem of manual participation in the processing process.This paper uses the current popular deep learning model BILSTM and CRF to combine the benchmark model,input the experimental data into the BILSTM model in the form of characters,tokenize the characters,and use the random vector field model to determine the dependence between the characters.Analyze,and finally get the optimal sequence label that accords with the realistic result.Experimental verification shows that geographic named entity recognition based on BILSTM and CRF models can effectively improve the accuracy and recall rate of obtaining geographic entity names from text data.(3)Aiming at the problem of relatively simplification of feature expression when using BILSTM to perform the task of geo-named entity recognition in the process of character-to-vector conversion,which cannot represent the accurate meaning of text characters.This paper proposes a geographically named entity recognition method combining the BERT model and the BILSTM+CRF model.The geographic text data is trained through the BERT model,the semantic vector is obtained from the contextual character information,and the BILSTM+CRF model is combined for recognition.Experiments have proved that this method has achieved 94.60% accuracy,92.46%recall and 97.52% F1 value in the process of geographically named entity recognition.Compared with the existing geographically named entity recognition methods,its effect is better,which proves the effectiveness of using deep learning methods for geographically named entity recognition.