A Study Of Several Key Problems In Construction Of Event Ontology

As a formal and explicit specification of shared conceptualisation, ontologyplays an increasingly important role in the application of artificial intelligencesuch as information processing and natural language understanding. At present,ontology is usually constructed as a relational system of various concepts, whichdoes not provide support for analysis in terms of time and space levels andcannot solve the “tennis problem”. Event ontology, taking event as the basic unitof knowledge, perfectly coincides with the way of human thinking. Moreover, itis able to express dynamic changes and provides a novel approach to overcomeobstacles of traditional ontology. The research on construction of event ontologyhas significant academic and practical value.In event ontology,“event” is not merely treated as static concepts orrelations between concepts. It is the basic unit of knowledge, which containsinformation like action, object, environment, as well as time. The concept“event” in event ontology is responsible for representation of relations betweendifferent events and relations between people or things involved in event.Furthermore, it should express the roles of event participants and dynamicprocedure of event. Based upon six-tuples expression of event, this paperfocuses on the key problems in construction of event ontology. Specifically, thecontributions of this thesis are listed as following:（1）Reuse of entities from existing ontologies and representation ofevent’s dynamic procedure based on concept algebra. In event, bothobject-element and environment-element are entities of existing ontologies. Toreuse those existing entities when constructing event ontology, the paper extendsoperations of concept algebra via adding “time flag” and “information quantity”.Representation of event’s dynamic procedure can be considered asrepresentation of a series of event’s states changing along with different time points, and then assertion-element can essentially express event’s states. Bytaking advantage of extended operations (inheritance, tailoring, extension andsubstitute) in concept algebra, we can change event’s states along with differenttime points. In short, the application of extended concept algebra in cognitiveinformatics greatly raises the efficiency when constructing event ontology.（2）Frame-based representation model of event and analysis of conceptualoperations. This paper proposes a novel frame-based representation model ofevent on the basis of six-tuples expression, analyzes several conceptualoperations, and defines the upper classification of events. Based on Nilsson’sconcept algebra, the frame-based representation model is able to express bothevent class and instance. Furthermore, it can express dynamic procedure ofevent and the relations between events and their elements. The frame-basedrepresentation model is flexible, effective and widely applicable. It is extremelyappropriate for representation of event-based knowledge and providestheoretical support for applications like knowledge storage and reasoning.（3）Event-based tagging and analytic technique for Chinese text. Buildingcorpuses of domain knowledge plays an important part in construction ofontology. Similarly, event-oriented corpuses tagging is essential for constructionof event ontology. In order to overcome shortcomings of existing Chineseanalytic technique and facilitate the build of event ontology, the paper presentsan event-based tagging and analytic technique for processing Chinese text. Incomparison with the tagging method of CEC (Chinese Event Corpus), itenhances the tagging ability and provides high quality corpuses for eventauto-recognition, auto-classification and retrieval of event elements or relations.The major extended functionalities are listed as following:①Classify and tagthe non-event components like preposition and auxiliary word.②Incorporateevents in text as much as possible③Ability of nested analysis for eventhierarchical structure.④Clearly denote event object, environment and time information from Chinese text.（4）Event auto-recognition and classification strategy. We retrieve8858event denoters from a bunch of text files, manually classify them to buildtraining set, and then design a one-against-one SVM (Support Vector Machine)learner with feature vectors containing word, lexical, grammatical and semanticinformation. Experiments indicate that accuracy of classification increases afterincorporating several effective features. By combination of some key features,especially semantic features, the SVM learner works effectively with theprecision up to81.85%. In summary, auto-recognition and classification of eventsignificantly reduce manual workload and lay a solid foundation for applicationlike event-based language understanding. By taking result from eventrecognition and classification, the matched event class with abundantinformation can be retrieved from event ontology, making it convenient forevent-based language understanding.