Search

This paper proposes a method to extract related news items with respect to a given piece of news from a collection of news items. The related news items include not only the news items with the same topic as the given one but also those implicitly related. In order to find truly related news items, the paper proposes a new query expansion method based on WordNet with the existing word co-occurrence searching method. The method can be used in event prediction and as a tool for information extraction. A benchmark data set with BBC news items is built up to test our idea. Experiments show a high precision and recall rate and a stable performance with different test news items.

A multiresolution method based wavelet is purposed for hierarchical document categorization. Multi-level wavelet decomposition of document vectors is regarded as a feature selection process. The traditional vector space model in document representation is reorganized by hierarchical conceptual-semantic relation of the words. The multiresolution wavelet coefficients of reordered vectors reflect the characteristics of the document category in hierarchy. At different levels of category, corresponding resolution wavelet coefficients are used to classify. The computational complexity of classification is reduced and better performance is obtained.