| Ontology,as an excellent knowledge representation method,has been widely used in recent years.There are many shortcomings in describing dynamic knowledge in traditional ontology models,such as the lack of describing dynamic changes between things and the problem of conceptual discreteness.Therefore,event ontology with event as its basic unit emerges at the historic moment.Based on event ontology with six elements,event ontology organically integrates participants,places and time in the event,avoiding the concept discreteness of traditional ontology.The emergence of event ontology overcomes the shortcomings of traditional ontology,and the related research of event ontology has opened up a new way of thinking and direction for the development of artificial intelligence.Among them,the construction of event ontology is one of the research hotspots of event ontology at present.The construction of traditional ontology is mainly based on manual construction,which is a complex work.With the increasing of the scale of ontology,the difficulty and complexity of construction and maintenance will increase dramatically,especially for event ontology.At the same time,as a new research direction in the field of ontology,the construction of event ontology lacks the corresponding construction methodology as the guidance of ontology development,and also lacks the supporting construction tools.In the development of event ontology,the core task is to obtain event classes and event elements in a specific domain,and to describe the relationship between events and events.This paper combines natural language processing and machine learning technology to automatically extract event and event relationships from text in a specific domain,so as to assist developers to quickly obtain candidate event classes and event relationships in the process of event ontology development,and ultimately achieve the goal of semi-automatic event ontology construction.On this basis,this paper also implements a Web-based semi-automatic visual event ontology construction platform,which supports multi-user and visual event ontology construction.The research work plays a positive role in improving the efficiency of event ontology development,and promotes the research of event ontology construction methods and the popularization and application of event ontology.The main contents and innovations of this paper are as follows:1.Chinese Emergency Recognition Based on Semantic Features and Neural Network: Event automatic recognition is one of the key technologies of semi-automatic event ontology construction,which is mainly used to extract event knowledge from text automatically.In this paper,event recognition is transformed into the classification of event feature vectors,and a classifier is constructed by using neural network technology.Firstly,candidate trigger words are filtered through emergency trigger vocabulary,and word sequences around candidate trigger words are selected as input of neural network.Then,an input sequence is constructed by combining word embedding features and semantic features.Then,a cyclic neural network and a classifier are constructed.The hybrid neural network model of convolution neural network transforms the input sequence into eigenvectors and classifies and recognizes them through the full connection layer.The experimental results show that the proposed method outperforms the traditional machine learning method in the Chinese emergency corpus.The research can be extended to automatic tagging of corpus and provide technical support for the large-scale expansion of emergency corpus in the future.2.Recognition of event causality based on twin networks: The construction of event ontology also involves the construction of event relationship.This work is also a part of the related research of event ontology construction.In this paper,we use the method of mapping event and event relationship to vector space to identify causality between events.We use word sequences around event triggers as input data,and use event-based twin loop networks to model events by encoding event representations into fixed size vectors,and then apply these event representations to relational embedding training and prediction.The experimental results show that this method can achieve better results for CEC 2.0 corpus.The first two research contents of this paper mainly serve for the research of semi-automatic construction methods and tools,mainly to realize automatic extraction of event-related knowledge from text to assist the construction of event ontology.3.Semi-automatic construction method and tool of event ontology construction: The existing ontology construction methods and tools mainly serve the construction process of traditional ontology based on concepts.The structure of event ontology is different from the stucture of traditional ontology,so these methods and tools are not applicable.Based on the seven step method proposed by Stanford University,this paper presents the event ontology construction method applicable to event ontology.In the process of ontology construction,event ontology pattern is introduced.Event ontology pattern is an event ontology model formed by extracting the commonness of event ontology in a certain field.Event ontology construction eight-step method avoids the repetitive work in ontology construction by reusing event ontology model,and improves the efficiency of ontology construction.In addition,we have developed a prototype of Web event ontology visualization construction tool,which was named Remaining Stone No.1,which is used to construct an air pollution event ontology by semi-automatic construction. |
PDF Full Text Request