Research On The Causality Recognition Method Of News Events Based On Cross-events

Information extraction has been widely attention by domestic and foreign scholars in recent years and the international community has set up a special ACE(automatic content extraction)evaluation conference to promote the development of the field.Event relationship is a very meaningful research content in information extraction,including event temporal relationship,event causality and so on.The news event causal relationship can help people understand the news well,grasping the evolution of news events.Event causality extraction is widespread used in the information extraction.In particular,it will have a wide range of applications in the automatic Q&A system.Hence the event causal relationship recognition has become a hot study site all over the world.The method of the existing event causal identification mainly uses local information that current events have and the recognition effect is not high.Other information contained in news reports is not fully utilized.This paper introduces the cross-event technology to identify the causal relationship of news events.When the local information is not enough to identify the causal relationship,we will actively search for other information in news reports.The corpus is the basis of the machine learning algorithm.This paper uses the Chinese evaluation corpus provided by the ACE2005 conference,which has been marked for entities,relationships and events.In this paper,a detailed analysis of the corpus is carried out,especially combining with the causal relationship recognition task.In addition,this paper also studies the causal relationship corpus construction and recognition fusing cross-event technology.Mainly the following characteristics of the work:(1)The Construction of New Event Causal Corpus.In this paper,Based on the analysis of the characteristics of ACE2005 Chinese corpus and the actual needs of event causality,the event elements such as event type,event polarity,event genericity,event trigger word and POS of event trigger word were extracted from the corpus.In addition,the personnel are organized to mark a news event causal relationship.Finally the results of the annotation will be stored to build a news event causal relationship corpus.(2)Identification of causal relationships of news events based on cross events.This paper bring cross-event technology into causal relation extraction,choosing feature space and construct maximum entropy classifier to finish sentence—level causal relation extraction.Setting a threshold value,keep the high probability part as the classification result,then construct document-level classifier to handle the part that with low probability.(3)Integration the results of the above research,design and implementation the prototype system of temporal relation identification.