Improving the referrer-based Web log session reconstruction

Web usage mining, an emergent domain in Web mining, is the process of discovering and interpreting patterns of users' accessing of the Web by mining the Web log data. We discuss the kinds of ambiguities a heuristic encounters during the process of user session reconstruction. Some criteria are introduced to measure these ambiguities. The proposed criteria are then used to adapt a reconstructor heuristic to the conditions of a particular server log. We extend the conventional referrer-based heuristic to exploit this adaptation in order to achieve higher performance as well as a reasonable stability. The results of simulation experiments are presented and analyzed in order to show the performance achievements and the stability tradeoff.