Study On Method Of Coreference Resolution From The Perspective Of Knowledge Graph

With the development of Internet,there comes the big data era,where the data collection by human beings has come up to the ZB scale.In order to make the query data more accurate and simplified,knowledge mapping has been widely applied to search engines.Actually,knowledge mapping is a relation network that describes the things in reality and their relationship such as the places,people,cities and movies.With knowledge mapping,it is possible for search engines to explore the inner relations among different objectives,to provide customers with precise information they need.Currently,the data of knowledge mapping mainly come from online encyclopedia,for example,Wikipedia,through automatic collection,which results in the problem of massive data and noise data.In this case,customers have trouble in acquiring the information needed and satisfying quickly.In terms of this problem,it is necessary to integrate the information collected.Therefore,to upgrade the knowledge mapping,what is worth being settled is how to make knowledge integration based on the current data accurately.Knowledge integration includes two parts,which are the entity liking and the knowledge combination.Entity linking contains entity disambiguation and entity resolution.Trough the entity disambiguation and entity resolution,it can be judged whether the entities with the same name in KB represent the same meaning,or entities with different names represent the same meaning.What attracts great attention now is the problem how to link the entity in the text and the entity in the KB quickly and precisely under the circumstance of limited contexts.According to the problem of massive and noise data of knowledge graph,this thesis studies the entity resolution in the process of knowledge integration from the aspects of knowledge graph construction technology as well as entity resolutionalgorithms analysis,We complete to construct the test platform of Chinese Coreference Resolution based on decision tree algorithm,oriented to the field of entertainment news.aiming to provide a new accurate and quick entity resolution algorithm.This article first uses the octopus collector to collect experimental corpora,and adds new features during the training classifier stage: contextual semantic features,Chinese abbreviation features,while improving the original distance features,using the sentence distance algorithm for feature calculation After repeated experiments,it was found that the added features have a good effect on the improvement of the classification accuracy.In the common-evidence stage,the above-mentioned classifiers can be used to more accurately find the items to be eliminated,and to eliminate the common indicators.As mentioned above,the thesis expects to offer a new algorithm so that customers are able to acquire the correct entity in short time and provide an effective method for search engines in the next generation.With the population and widespread of knowledge mapping,the study of this thesis can be applied in various fields such as business,finance and life science.It will be helpful to query information on business judgment,financial analysis and biological medicine product,which makes great contribution to the society development.