A Model for Clustering Clickstream Data

The extremely large number of data sets that can be drawn from internet has bootstrapped in a way the data mining techniques from extracting real time results. For this reason clustering and other mining techniques in the data stream model have grasped the interest of Data Mining community [6]. Recently have been proposed many algorithms for the basic clustering problem for massive data sets [7] that produce an approximation using efficiently the memory, which is the most critical resource for streaming computation. Also have been studied some approximation algorithms [8] for the k-median clustering using corsets in Rd with only polynomial dependency on d. In this paper, we present a new model for clustering click stream data. We adopt the idea of division the clustering process into an online component which periodically stores detailed summary statistics and an offline component which uses only this summary statistics [3]. In this concept we use k-means in order to perform clustering over data streams which is a very efficient and powerful algorithm for this type of problems. For the purposes of our experiments we have modified k-means in the input process. In other words we have built a mechanism that provides k-means with real time data feeds from web server. The main originality of our model is based on the fact that it uses compressed data of click streams in memory providing on-line clustering with real time results. These results allow further cluster analysis and have security and usability perspectives