Research On Event-based News Story Analysis Technology

News story is the important carrier of information. In everyday work, lots of information is recorded, delivered and exchanged through news story. When facing these tide-like electronic references, people are eager to have all kinds of intelligent services, which can automatically collect, filter, organize and utilize network information. Event-based news story analysis is a powerful tool, the aim of which is to effectively organize and process vast news information. Therefore, this dissertation explores the topic on event-based news story analysis technology, which is a research issue with great significance in theory and wide perspective in application. The main achievements in this dissertation can be described as follows:●A frame of event-based news story analysis is proposed. Technology frame is constructed based on the concept analysis and the hiberarchy of event-based news story analysis technology. The definitions of relevant terms are analyzed in concept analysis and the hiberarchy of event-based news story analysis technology is also presented. The technology frame discusses the approaches to realize event-based news story analysis based on the concept analysis and the hiberarchy and points out the problems this dissertation concentrates on.●Some news event detection methods are proposed and improved. Event detection method based on incremental K-means is firstly improved. It is known that traditional incremental K-means cannot select initial cluster centers objectively in event detection. To solve this challenging problem, this dissertation utilizes density function to initialize cluster centers. The problems of effective density radius selection and feature space dimension selection are also discussed in this dissertation. An ICURE-based method for performing the event detection is also proposed, the algorithm can solve the problems in computational complexity and data updating effectively.●Some news event tracking methods are proposed and improved. A method based on K Nearest Neighbor Feature Line (KNNFL) was proposed for tracking events. NFL combining with improved KNN produces KNNFL in order to make it more suitable to news event analysis. A negative-examples-pruning support vector machine (NEP-SVM) based algorithm is also proposed for event tracking. It reserves and deletes a negative sample according to distance and class label, Finally, the SVM outputs are mapped into probabilities.●An event relevant multi-document summarization method is proposed. The specificity of the method is that the basic-partial-topic-sentence group (BPTSG) and the extended-partial-topic-sentence group (EPTSG) are generated based on the extraction of the basic news factors and the extended news factors, which can both cover as many themes as possible and reduce its redundancy at the same time.●An event-based news story analysis system is designed and implemented. The design idea and each functional module of system IEventMiner are described in detail, and the implementation of prototype system is also presented, which gives a sound support to the frame and relevant methods of event-based news story analysis.As a general, the dissertation focuses on the key techniques of event-based news story analysis, such as event detection, event tracking, event related multi-document summarization and so on, consequently made progress in some domains. The designed system IEventMiner uses the design of module structure, so it is easy to extend and improve performance. Not only does this dessertation promote the development of system engineering, but also it can be referred by intelligence analysis.