Research On Event Recognition And Its Application For Unexpected Events

Events are not only the basic unit of knowledge for people to carry out memory knowledge and logical thinking,but also the language for information dissemination and information preservation.As the bond of realistic experience and linguistic signs,events have received more and more attention in recent years.Event recognition is an important part of event research.Its role is to identify the words in a sentence that represent the occurrence of an event.In the research of present event recognition,however,exist the problems of solely selecting feature and vulnerable recognizing model caused by irrelevant redundancy feature.Therefore,a multi-feature event recognition method is proposed in this paper,and on this basis,a research targeting the classifying application of unexpected events was carried out.The main research work includes two parts:1.Research on event recognition for unexpected events.The current research on event recognition relies primarily on one or more features like semantics,syntax and so on,and ignores the irrelevant and redundant characters.Nevertheless,an unexpected event includes quite many trigger words in a limited sentence.It is necessary to find more effective features from a limited vocabulary to distinguish event trigger words.So this paper proposes a multi-feature event recognition method to deal with the above problems.The main features like lexical features,semantic character features,dependent syntax features,and semantic dependency features are firstly selected.Then,the Relief algorithm is used to select sub features,unrelated features and redundant features to deal with the issues above.Finally,the feature vectors after feature selection are classified by the SVM classifier to realize event recognition.Experimental results show this method works well.2.Research on application of text topic classification for unexpected events.This article divides it into two parts,a training phrase and a testing one.The first one is to process the event annotation results in the corpus,value all event trigger words for each topic,and get the classification template for each category.The other one compares trigger words one by one from classification and identification,and value them again.In that way,it can calculate the value of each trigger word in the text.And add them up as a relative sum of weight in turn.The texts to be classified are calculated sequentially relative to the total weights of five categories.Then,choose the classification,with a largest weight,as a final result.Experimental results show that works well,too.