| Event is the basic unit of knowledge representation,when human beings communicate with each other.The view that human beings do thinking activities with the unit of event has been approved by more and more researchers.Since the MUC proposed event extraction from text,it has attracted more and more attention.Especially recent years,event extraction from text is one of the researching hot interests in NLP field.Nowadays,traditional text minding methods mainly use words,concepts or phrases as unit to represent knowledge.However,the traditional text mining method has some shortcomings,such as,lacking of semantic information,tennis problems,without abilities to express and reason high-level semantics and so on.From the aspect of event,a text,especially a narrative text,to a certain extent,is a kind of written expression about people's understanding on a series of events and event relationships in the real world.Event is treated as unit to represent semantic knowledge of text,which can't only solve the shortcomings of the traditional text mining methods,but also represent and reason high-level semantic knowledge of text.In recent years,the research on text event mainly focus on event extraction and some event-based applications.And the knowledge about text event is the basis for them.Therefore,we treated event as the basic unit to represent knowledge of news texts in the Internet,and researched on mining and representing knowledge about text event.We wish our work benefits to semantic knowledge understanding,constructing event ontology and some event-oriented applications.Our main work and contributions are as follows.(1)Event-based text optimization annotation and statistics: On the basis of CEC 1.0,this paper optimized and supplemented the event-based annotation specification,including idea event,event relationship and the tool or method that describes the action of event,and so on.According to the specification,annotated these semantic information and constructed the CEC 2.0 corpus.Additionally,this paper made statistics and analysis for C EC 2.0 from the number of texts,event factor and annotation effect etc.the results show that the CEC 2.0 doesn't only increase the number of annotated texts from 200 to 333,but also have more abundant semantic information.The CEC 2.0 is superior to the CEC 1.0.(2)Discovery association rules and collocation patterns for core words of event language expression: this paper proposed the method that Apriori-based core word association rules discovery for event language expression,The method treated each event in C EC 2.0 as a transaction and the core word and its positional feature,part of speech as items of the transaction,and then mined the association rules with Apriori algorithm.Additionally,this paper proposed a method for mining the core word pattern based on semantic dependency analysis.Firstly,this paper did semantic dependency analysis for events in the C EC 2.0 corpus,then processed the semantic dependency tree,and then mined,finally,mined the core word collocation pattern with PETreeMiner algorithm.The example verifies that the mined association rules and collocation patterns can help automatically generate sentences,which indicates that the proposed is valid.(3)Knowledge representation and reasoning about event and event calss semantic: this proposed a method for representing event and event calss semantic by combining new Davidson and 6-tuple event model.It treated event predicate as a singleton that contains only event argument,and combined it with the six factors of event by ?.This paper extended some operators to represent action,object,tense,environment and event relationship and represented the concept in object with Description Logic.Based on these,this paper proposed the reasoning methods for the missing factors of event and the follow up event.The methods treated the represented event,event class and some rules as knowledge base,and then did reasoning with the rules.The case of verification shows that the proposed methods have good abilities to represent knowledge and reasoning. |
PDF Full Text Request