Named Entity Recognition And Relation Extraction Research And Application

With the improvement of informationization,Internet information increases quickly.How to quickly and accurately extract the content of interest from the vast amount of Internet information has become an urgent problem to be solved in the field of natural language processing.The main purpose of information extraction is to convert unstructured text into semi-structured or structured data to facilitate rapid retrieval of information by users.Named Entity Recognition and Entity Relationship Extraction,as the core task and important part of information extraction,plays an important role in semantic understanding and knowledge graph construction.This paper mainly studies the technology of named entity identification and entity relation extraction.The main research results are as follows:(1)An end-to-end named entity recognition model of BiLSTM-CNN-CRFs based on deep learning was constructed.Solving traditional machine learning is difficult to obtain shortcomings such as long dependencies and labor costs.This method uses the CNN layer to perform feature extraction and selection on the input sentence,and uses CRF at the output end of the LSTM to decode to form the best tag chain of the text sequence.Experiments show that this model can achieve better recognition results.(2)This paper proposes a two-stage entity relationship extraction method based on three-way decision.Aiming at the severe overlap of SVM classifiers in the vicinity of the hyperplane in the multi-class scenario,The SVM three-way decision classifiers are constructed to implement the first-phase entity relationship extraction.The softmax multi-classifier function is adopted as the three-way decision probability functions.Then the KNN classifier is used to classify the three intermediate domain samples after the decision-making.Taking the corpus of ACE2005 as the experimental data,the selected words,parts of speech,physical location information and entity type are taken as the features.The experimental results show that the two-stage entity relation extraction method based on three-way decision makes a good extraction effect.(3)A domain named entity recognition method based on feedback K-nearest-neighbor semantic transfer learning is proposed.Firstly,the corpus of professional fields and general fields are respectively trained to obtain corpus document vectors.For each field of specialization,K samples with the most similar semantic fields are selected for semantic migration learning,and N migration corpus sets are constructed.Then,the BiLSTM-CNN-CRFs network model is used to identify domain-named entities in N migrating corpus sets,and the recognition results are evaluated and fed-forward.Based on the feedback results,an appropriate K-value is selected as the optimal threshold for semantic learning.The results show that this method has achieved good recognition effect and can effectively solve the problem of lack of corpus in the field of specialization.