Research On Chinese Named Entity And Entity Relationship Extraction

Named-entity and extraction of the entity relation have always been the basic and key subtasks in the task of information extraction,and at the same time the key basic link in the comprehensive applications such as intelligent Q&A and machine translation in the field of natural language processing,which have become the subject that scholars at home and abroad have been widely discussing and deeply studying in recent years.In the face of the challenge of seas of unstructured and semi-structured text data brought by the rapid development of Internet,problems such as the lack of large-scale standard corpus,the unintuitive of output,the time-consuming and energy-consuming artificial feature extraction and long-distance context features need resolutions.It is innovatively put forward in this paper to use the methods of co-occurrence matrix and sliding window to solve the problem of long-distance context features,and to use multilayer encoders to extract word features to solve the problem of time-consuming,energy-consuming and domain-knowledge-needing artificial feature extraction,in the meantime enhancing the generalization ability of models.Problem extraction is ed into problem classification to make the final output intuitive and reliable.It is indicated through the text results based on authentic materials that the extraction result by the innovative extraction method put forward in this paper,compared with that by the method based on maximum entropy model and feature fusion,is averagely 9.3%higher in precision,5.5%higher in recall rate and 7.5%higher in F value.From the result of comparison,deep neural network model not only can apply to the extraction of named-entity in Chinese,but also is a big increase in the extraction performance compared with shallow statistical learning model.Study in such aspect will provide theoretical and practical basis for using the theory and technique of deep learning to process natural language problems.