Research On Key Technologies Of Event Extraction On Event Evolutionary Graph

An event is not only the basic component parts of knowledge for people to gain knowledge and logical thinking,but also a way of dissemination and storage of knowledge.In the computer field,logic,an evolution rules and patterns between events,is a common sense.It is extremely essential to find out the changing regulation happening to understanding human behavior and the development of society.The event graph is not the knowledge base of nouns as the core node,but the logical knowledge base of abstract events as the core node.As the most important part of the research of the event graph,the event extraction is to extract the needed information from texts.The traditional event extraction manly includes the event recognition and event element recognition,but it is not enough for event extraction for event knowledge graph.This thesis provides a new event description by divide events into main events and subordinate events meanwhile provides a new way to extract information for texts.This thesis aims to optimize the event extraction algorithm and put it into real usage.The research are included in the three followings:1?The fast labeling based on double pattern matching: Event knowledge map event extraction uses supervised deep learning algorithm in the use of the algorithm,so it needs to use tagged data,while manual tagging method requires a lot of manpower and time.This paper proposes a fast annotation method based on double pattern matching,i.e.manual rules and rules based on labeled corpus acquisition,which improves the efficiency of labeling and reduces the manpower and time required for labeling.2?Dynamic word vector representation mixing the context dependence and sentence semantics: the word vector provides a numerical method of textual data.The traditional word vector has mostly uses the methods of static word vector.Accordingly,the thesis employs the method that can express polysemants in different sentences through different dynamic word vector.It also offers comprehensive word vector for the event extraction,which improving the accuracy of the models.3?Design of Event Extraction Algorithms: The event extraction algorithm of event knowledge map is proposed.The event extraction of event knowledge map is defined as three tasks: event partition,event element recognition and event classification.The three tasks of event extraction of event knowledge map are experimented and validated by using neural network model.However,the neural network model ignores the traditional features,but its design and verification by experts is effective.Therefore,this paper proposes a method that combines the neural network with traditional features to optimize the model and improve the effect of the algorithm.In conclusion,this thesis has accomplished the related design and proved the practical usage the feasibility of innovative solution through massive contract experiments.