Resourc Entity And Relation Extraction For Free Text

The resource database is an important database in the Computer Aided Innovation System,which has been filled for years by manual extracting resource from mass information.The amount of data in the resource database is of great value to the CAI.Therefore,how to automatically and efficiently extract resource from massive texts has attracted attention and research.The method of machine learning is often adopted to extract resource.This kind of machine learning model is usually a shallow model.Features are developed around two entities and remain on the lexical surface.However,resource extraction needs to be based on the semantic information of the whole sentence,so it is very important to obtain the deep semantic features of the whole sentence.Therefore,this paper proposes to extract resources by using the Bidirectional Long Short Term Memory.This model has the characteristics of mining the deep semantic information of text and using the sentence context information,which can obtain better experimental result.Therefore,a resource entity and relation extraction algorithm is proposed.The purpose is to extract the substance,the attribute,the parameter,the value in the free text and to extract the relation between them.The algorithm includes resource entity recognition and resource relation extraction.In the process of resource entity recognition,because the attribute and the parameter are in the same syntax position in a sentence,it is difficult to distinguish them.So the attribute and the parameter are treated as an entity,called the attribute-parameter entity.The entity recognition model is constructed by BLSTM for the initial entity recognition.The recognition results are corrected by the combination of dictionaries and rules.In the process of resource relation extraction,the relationship classification model is constructed by BLSTM for determining whether there is a relationship between the attribute-parameter entity and the value entity.Then a method of combination of dictionaries and rules is proposed to distinguish the attribute entity and parameter entity,and to deal with the problems of the value entity that do not match the attribute-parameter entity.Through this algorithm,<substance,attribute,parameter,value> is finally formed to fill the database.Finally,this paper makes the standard of corpus annotation.A resource corpus is formed to carry out the experiment of resource entity recognition and resource relation extraction.Both experiments have achieved satisfactory results.By contrast experiments,BLSTM is proved to be excellent in entity recognition and relation extraction.