An adaptive topic tracking approach based on Single-Pass clustering with sliding time window

Topic Tracking is one hot branch of Topic Detection and Tracking (TDT) research field. Due to the sparseness of initial corpus, the traditional topic tracking methods usually bring topic excursion into the system. The adaptive topic tracking methods are being put forward, but most of them use the fake feedback strategy so that the performance is not improved too much. In this paper, we proposed an adaptive topic tracking approach, which is based on an improved Single-Pass clustering algorithm with sliding time window. We present our own corpus preprocessing and feature weighting method to provide more exact vector space model (VSM) to next stage. In the topic tracking process, this paper uses a sliding time window strategy to guarantee the system accuracy and reduce the number of missed following stories. The experimental results show that our approach achieves satisfying results.