Named Entity Recognition Of The Code For Geology Investigation Of Railway Engineering

As a basic task in the field of natural language processing,named entity recognition is a key technical means for mining and managing massive text information,and plays a key role in the construction of knowledge map.In the field of Code for Geology Investigation of Railway Engineering,because the existing universal named entity recognition model mainly focuses on the research of universal data and lacks the support of the geological field corpus,the research work is basically blank.In this paper,based on the relevant text of the Code for Geology Investigation of Railway Engineering,the corpus of the Code for Geology Investigation of Railway Engineering is constructed,and the named entity recognition is studied by using the constructed corpus.The main work is as follows:(1)Construction of geological field corpus: referring to the national geological industry standards and specifications such as TB 10012-2001 · J 124-2001: Code for Geology Investigation of Railway Engineering,TB 10027-2001 · J 125-2001:Code for Unfavorable Geology Investigation of Railway Engineering,two small-scale entity identification corpora of railway engineering geological survey specifications have been designed and constructed for the first time,and two data sets in the field of railway engineering geological survey have been generated.(2)Construction and experiment of named entity recognition model in the geological field: The Bert-based pre-training model BERT-BILSTM-CRF model was used to carry out entity recognition tasks in the geological field.Experiments were conducted on three general data sets and the geological field data set Goe NER2021,and the prediction effects of the four models on entity recognition tasks in different types of data sets were analyzed and compared,and Proved the superiority of the algorithm.(3)Model re-evaluation: based on the data set in the field of Code for Geology Investigation of Railway Engineering,the prediction effect of the optimized model is evaluated again,and the engineering type of the first-level entity,the characteristics of the geological conditions and the recognition results of the secondlevel entity are further analyzed according to the predicted output results,and the unmarked text is predicted and analyzed,further improved the capability of the model.