Web page prediction enhanced with confidence mechanism
Árpád Gellért, Adrian Florea · Journal of Web Engineering · 2014
In this work we comparatively present and evaluate different prediction techniques used to anticipate and prefetch web pages and files accessed via browsers. The goal is to reduce the delays necessary to load the web pages and files visited by the users. We have included into our analysis Markov chains, Hidden Markov Models and graph algorithms. We have enhanced all these predictors with confidence mechanism which classifies dynamically web pages as predictable or unpredictable. A prediction is generated only if the confidence counter attached to the current web page is in a predictable state, improving thus the accuracy. Based on the results we have also developed a hybrid predictor consisting in a Hidden Markov Model and a graph-based predictor. The experiments show that this hybrid predictor provides the best prediction accuracies, an average of 85.45% on the Ocean Group Research dataset from the University of Boston and 87.28% on the dataset collected from the educational web server of our university, being thus the most appropriate to efficiently predict and prefetch web pages.