Research On Key Technology And Application Of Event Generalization

Traditional Eventic Graph mostly focuses upon the specific event information such as transfer types of event and event attribute,while it ignores the evolution laws and logical relationships of reasons themselves.In order to fully make up for the deficiency,many scholars majored in related field put forward the concept of event generalization in the work associated with events.At this stage,the development of semantic network technology has greatly improved the effect of event relationship related tasks,and also made it possible to mine the potential rules and logical connection between events through the text.Starting from the unsupervised generalization of events,this papers stresses on the exploration and studies of the abstract event network,including the unsupervised generalization of events,the construction of the event network of abstract events and the prediction of future events based on the above work.One way to explore the logical relationship between events is event generalization.Highly accurate event generalization is the basis of subsequent task application.This paper mainly explores at the direction of unsupervised event generalization which based on the explicit semantic relationship in semantic network via the loss of strong constraint function constrained on the event argument in the generalization process.In order to make sure that the generalized abstract event pairs still maintain the original event relationship type,this paper further calculates the causal strength between abstract event pairs through the pre-trained model to filtrate the abstract event pairs found no meaning after generalization.Finally,experiments are carried out on two causal event datasets forming in Chinese and English,and a large number of high-quality Abstract events are obtained in practice.Based on the above events,this paper explores the event network based on hierarchical clustering by means of clustering initialization which reduces the construction complexity and constructs a dense event network.We also further integrates abstract events and event networks through the knowledge fusion method,aiming to vastly integrate more generalized information into the map.Experiments on semantic network data sets show that the knowledge fusion model has the ability to effectively fuse abstract events to hierarchical event networks.Event prediction is one of the crucial application directions of event knowledge.This paper seriously explores an event prediction algorithm based on the graph pattern,which makes fully use of cluster centroid matching method to predict events which most probably occur.According to the experimental results,compared with other prediction algorithms without training,the event prediction algorithm based on centroid matching has received a substantial increase of improvement;Compared with the event network without abstract event nodes,the event network fusing with abstract information offers better support for downstream tasks.