Knowledge Extraction By Sentence Level For Academic Papers

This dissertation first makes an analysis on the syntactic structure and semantic structure of new sentences from academic literature,and both structures are identified through similarity between sentences.Then it analyzes relations among sentences and discourse structures of documents so as to judge function of various knowledge elements contained in sentence,including definition,attribute,methodology,data,result and conclusion.Knowledge extraction based on sentence match and analysis can not only achieve citation index automatically by identifying similar sentences,and help write literature reviews by clustering new sentences,but can also organize knowledge on the level of knowledge elements.This study analyzes document content and structure with multi-rules on sentence level rather than on lexicon level.Knowledge extraction aims to build content metadata about definition,classification,term attribute,methodology,result and conclusion based on discerning paper type,style and syntactic structure.The dissertation selects different knowledge extraction modes depending on different document content metadata,for instance,definition extraction could be realized through linear presentation of sentences,while classification extraction under the aid of thesaurus and ontology.This dissertation approaches knowledge extraction based on sentence from a few perspectives:basic theory,methodology,key technology and experimental analysis,then presents a knowledge extraction platform,which can realize some applications,such as definition extraction,method extraction.The dissertation makes an experiment,as a test,to extract knowledge about research methods from the full-text papers published in the Journal of the China Society for Scientific and Technical Information in 2012.The novelty or value of this dissertation is reflected in the following aspects:(1)The dissertation studies system architecture,processes,and methods of sentence-level knowledge extraction,and uses a small-scale data sample to test the feasibility of proposed knowledge extraction methods.Experimental results show that the method of knowledge extraction based on multi-stage rules is effective.(2)This dissertation improves the accuracy of the knowledge attribute determination through the two sets of rules---forward rules and inverted rules.By using such rules,the definition,processes,classification,characteristics and the function of the research methods could be teased out.