A Query Language and Its Processing for Time-Series Document Clusters

摘要

Document clustering methods for time-series documents produce a sequence of snapshots of clustering results over time. Analyzing the contents (topics) and trends in a long sequence of clustering snapshots is hard and requires efforts since there are too many number of clusters; a user may need to access every cluster or read every document contained in each cluster. In this paper, we propose a framework to find clusters of user interest and change patterns called transition patterns involving the clusters. A cluster in a clustering result may persist in another cluster, branch into more than one cluster, merge with other clusters to form one cluster, or disappear in the adjacent clustering result. This research aims at providing users facilities to retrieve specific transition patterns in the clustering results. For this purpose, we propose a query language for time-series document clustering results and an approach to query processing. The first experimental results on TDT2 corpus clustering results are presented.