The Research On Event-Oiented Text Representation

Event as the unit of knowledge, can express the higher granularity semantic meaning than static concept, which has received more and more attention during these years. Efficient processing and apply text is necessary because text is the information carrier and is an important vehicle of knowledge transmission. The various texts such as novel, play, biography, news, which contain many events. From the view of semantic understanding, linguists think text semantic is a set of not only attributes and concepts, but also events and inter-event relations in higher granularity. Event-oriented text representation, which conforms to human reading and memorizing laws, and expresses semantic information in a higher level, and provides new methodological basis for semantics-based knowledge processing.In recent years, there is a tendency to use richer text representation than just keywords and concepts in text processing. Aiming at the drawbacks of conventional text representation, event network: a novel semantic-based and event-oriented text representation model is proposed in the paper, and event network can save semantic information of text, reflect events and inter-event relations, and also embody the information of the event importance and dynamic behaviors of event. It is necessary to study the properties and operations of the model.The operations on event network are good forms of semantic computation, which can provide support for text semantic information processing. The content and innovation of this paper include:(1) Event similarity calculation. The similarity between events is defined according to a given event model. The similarity between event elements as the indicator of similarity between events, event similarity calculation is changed into event element similarity calculation, which combines syntax, semantic, word sequences and time relation according to the features of event elements.(2) Research on structure and property of event network. The structure and the representation of event network were defined, and event network system was stated. The event network of text is built by taking event as text feature, the inter-event relations as edges between event nodes. The experiment shows that the event network has the small-world properties. Based on the conclusion, important events are selected according to the event importance to make the event network small, and these important events can make a good summarization of the text.(3) Matching for event network. It is necessary to investigate the appropriate method for the comparison of two texts in any of these text representations. Most of matching theory come from graph theory and information retrieval, graph matching is an important part of graph theory, conceptual graph matching is a central part in information retrieval, however, there is a significant difference between event network and graph and conceptual graph both in structure and semantic. The existing matching methods can not be used directly for event network matching. Therefore, it is necessary to study a novel method for matching event network. The paper presents a new method for event network matching: event matching based on maximum similarity priority, relation matching based on isotropic-relational-distance matrix, the third and final step is event network matching through integrating event matching and relation matching.