Research On Recognition And Reasoning For Event Plot Relations

Story is made up of a series of events.The plot structure is the overall design or layout of the events in story.It is an intuitive idea to structure events with plot struc-ture in narratives,because humans tell stories in this way.This paper focuses on the connecting relations between events in plot structure,event plot relations.Event plot relations are explanatory relations between events,which can represent the meaningful connecting relations between events in narratives.Recognition of event plot relations can benefit lots of Natural Language Understanding tasks,such as summarization,ques-tion answering systems,and co-reference resolution,among others.Recognition of event plot relations is a new research topic.To the best of my knowledge,there is little previous work on the recognizing event plot relations auto-matically.This study focus on the approach to automatic recognition and reasoning for event plot relations.The main contents are as follows:(1)Event plot relations recognition approach based on local predictionThis paper proposes a event plot relations recognition approach based on local pre-diction,which extracts several kinds of features to train a machine learning model predicting the categorical probability distribution of event plot relations.The results of experiments show that the probability prediction performance of this approach is better than baseline.(2)Event plot relations recognition approach based on global optimizationBased on the local predicted results of event plot relations,this paper proposes an integer linear programming model to optimize the event plot relation problem glob-ally.This approach designs several basic constraints and anti-transitive constraint for event plot relations,and obtains global consistent optimization results.The re-sults of experiments show that this approach improves the performance of event plot relation classification.(3)Joint reasoning for event plot relations and event elementsFinally,this paper proposes a joint reasoning approach for event plot relations and event elements.On the basis of the relationship between event elements and event plot relations,this paper designs two constraints in this approach on event partici-pant elements or event location elements numbers limit and joint reasoning.Based on the framework of the global optimization approach for event plot relations,this approach achieves multi-tasks optimization results by redesigning object function and adding multiple constraints.The experimental results show that the perfor-mance of this approach is significantly improved.In addition,there is no need to make extra annotations after local prediction models is trained so that this approach has practical value.