Named Entity Recognition Of Middle School Mathematics Knowledge Based On Deep Learning

Named Entity Recognition(NER)is a basic downstream task in the field of natural language processing.Its purpose is to identify entities such as persons,locations,and organizations indicated in text corpus information.NER plays an important role in various natural language processing applications,such as text understanding,information retrieval,text summarization,machine reading,machine translation,etc.As a downstream terminal task,the accuracy and speed of the recognition of named entities of middle school mathematics knowledge will directly affect the subsequent machine understanding.The machine article aims at the semantic cross-decomposition existing in the named entities of middle school mathematics knowledge,complex logical relationships and high recognition in a short time.The main tasks and innovations of precision requirements and other issues are as follows:(1)Propose a BERT-BILSTM-CRF-based named entity recognition model for middle school mathematics knowledge.This model combines the BERT(two-way long and short-term memory model and conditional random field,two-way long and shortterm memory model and conditional random field)model.Using BERT for pre-training,creatively proposed to enrich words based on the contextual semantic information of the text instead of traditionally replacing the correlation with these sequences.The correlation can be learned through the CRF layer.The composite model improves the mathematics of the middle school.Accuracy of knowledge of named entity recognition.The model is verified,and the F1 value is 26.25 substitutions and 16.1 substitutions higher than a single model such as CRF and BILSTM,which is 7.56 subdivisions higher than the F1 value of the hybrid model BILSTM-CRF,and is higher than the F1 value of BERT-CRF.0.91 points higher.(2)In order to recognize the limitation of BERT-BILSTM-CRF in training rate,a new middle school mathematical knowledge named entity recognition model based on BERT-BIGRU-CRF is proposed.This model first obtains the word vector through the BERT pre-training language model.Use the BIGRU(Bi-Gating Recurrent Unit,twoway gated recurrent unit)network to obtain the overlapping features of the insertion distance in the text,and finally obtain the best replacement sequence through CRF model decoding.The BERT-BILSTM-CRF model can obtain better accuracy than the BERT-BILSTM-CRF model in a limited time.Within the learning rate of 0.001-0.01,the F1 value of BERT-BIGRU-CRF is greater than or equal to BERT-BILSTM-CRF.