How to Perform Incremental Clustering - A SOM Based View

Due to fast development of network technique, internet users have to face to massive textual data every day. Because of unsupervised merit of clustering, clustering is a good solution for users to analyze and organize texts into categories. However, most of recent clustering algorithms conduct in static situation. That indicates, it doesn´t allow clustering algorithm to deal with novel data efficiently. When novel data appear, traditional clustering algorithms can´t change their structure easily. Obviously, this restrict is not fit to internet, since novel data appear at any time. For this reason, an incremental clustering algorithm is proposed in this paper to cluster incremental data. This algorithm has two factors. (a) It designs two measures to calculate feature´s ability and integrate them in similarity measurement by replacing concurrence based similarity measurements. (b) Based on proposed similarity measurement, this algorithm selects few samples from original texts to perform incremental clustering. Experimental results demonstrate that, after integrating feature´s capacity, our algorithm can obtain high quality to cluster texts.