Using Similarity Measure to Enhance the Robustness of Web Access Prediction Model

Prefetching web content by predicting users’ web requests can reduce the response time of the web server and optimize the network traffic. The Markov model that is based on the conditional probability has been studied by many researchers for web access path prediction. The prediction accuracy rate can reach up to 60 to 70 percent high. However a drawback of this type of model is that as the length of the access path grows the chance of successful path matching will decrease and the model will become inapplicable. In order to preserving the applicability as well as improving the accuracy rate, we extend the model by introducing a similarity measure among access paths. Therefore, the matching process becomes less rigid and the model will be more applicable and robust to the change of the path length.