Construction And Application Of Event-driven Event Knowledge Graph

The traditional knowledge graph focuses on entities and relationships between entities.Entity is a static concept,while event is dynamic,which contains richer semantic information,and the granularity of expression is greater than entity.With generalized event knowledge as the main content,event knowledge graph can better describe the development process of occurrence and evolution of events,and clarify the rules of occurrence among events.Based on the CEC2.0 Chinese Emergency Corpus,this dissertation acquires event class knowledge and event class relationship knowledge,and constructs event knowledge graph.And use the rule method to carry out the follow-up event prediction work.Research lays a foundation for event-oriented knowledge discovery,processing and application,and can also use the event evolution law in the event knowledge graph to provide support and basis for emergency emergency decision-making,so as to reduce the loss and injury of people's property and life caused by emergency.The main research contents of this dissertation are as follows:(1)Acquisition of event class candidate trigger words.A word similarity calculation method based on synonym lexical forest is proposed.This similarity calculation method can be used to cluster event trigger words in corpus effectively,so as to further obtain event class candidate trigger words.(2)Acquisition of event class object elements.A method of finding event class object elements based on Big Ci Forest is proposed.Big Ci Forest is used as the knowledge base of mapping its objects to the upper concepts.(3)Acquisition of environmental elements in event classes.A method of finding environmental elements in event classes based on dependency syntax analysis is proposed.The core words of environmental elements are determined by dependency syntax analysis.(4)Acquisition of event class relations.A method of co-occurrence of events is proposed to judge the correlation of event classes,and the occurrence probability of event classes with correlation is calculated.Then,based on the annotated event relations in the emergency corpus,the event class relations are mapped to the event class relations to realize the discovery of emergency class relations.(5)Visualization and application of event knowledge graph.The event classes are taken as nodes,the relationship between event classes as edges,and the probability of occurrence as weights.The event knowledge graph is visualized,and the constructed event knowledge graph is taken as knowledge to conduct subsequent event prediction research in a rule-based method.Experimental results show that the event knowledge discovery method proposed in this dissertation is improved compared with other methods in terms of accuracy P,recall rate R and F1 indexes.Experimental results show that the proposed knowledge discovery method of event class relationship is effective.The constructed event knowledge graph lays a foundation for future event-oriented theoretical research,and also provides a factual basis for making emergency decision plans,so as to control or reduce the damage caused by emergencies.