Semantics-based event-driven web news classification

Web news classification is an unsupervised learning task, which is often accomplished by clustering methods. In traditional works, documents are first represented using the vector space model. Each vector generally consists of the keywords or phrases important to the document. Then vectors are clustered together according to some (dis)similarity measure. Such methods often take no or little semantic information into account. In this paper, we present a semantics-based event-driven approach. Event is represented by 3-tuple and document is associated with set of candidate events. These event sets are classified according to semantic dissimilarity. The preliminary experiment on Chinese web news classification shows that the proposed approach is promising.