This paper studies the problem of location-based timeline generation from a syndicated stream of textual documents. We propose a framework containing two parts:location-based stream clustering and timeline generation. The first deals with the syndicated stream. Given a document from a syndicated stream, a technique named geo-tag is proposed to geo-focus each document up to POI-level. Next, we build a page model that considers both geographical and temporal feature, and get a series of geo-tagged clusters. Finally, we propose an approach to extract hot topics among the syndicated documental stream, and obtain a stream of geo-tagged hot topics continuously. The second performs range/continuous timeline computation relying on a model called hot-sensitive page rank. We also propose a technique called divisive top-K document selection to expedite the timeline computation.The experiments on real datasets demonstrate the effectiveness and efficiency of our proposed methods. |