Machine learning methods for one-session ahead prediction of accesses to page categories

Authors: Martín-Guerrero, J.D., Balaguer-Ballester, E., Camps-Valls, G., Palomares, A., Serrano-López, A.J., Gómez-Sanchís, J. and Soria-Olivas, E.

Volume: 3137

Pages: 421-424

DOI: 10.1007/978-3-540-27780-4_65

Abstract:

This paper presents a comparison among several well-known machine learning techniques when they are used to carry out a one-session ahead prediction of page categories. We use records belonging to 18 different categories accessed by users on the citizen web portal Infoville XXI. Our first approach is focused on predicting the frequency of accesses (normalized to the unity) corresponding to the user's next session. We have utilized Associative Memories (AMs), Classification and Regression Trees (CARTs), Multilayer Perceptrons (MLPs), and Support Vector Machines (SVMs). The Success Ratio (SR) averaged over all services is higher than 80% using any of these techniques. Nevertheless, given the numerous quantity of services taken into account, and the variability of SR among them, a balanced performance is desirable. When this issue is analysed, SVMs yielded the best overall performance. This study suggests that a prediction engine can be useful in order to customize user's interface. © Springer-Verlag 2004.

Source: Scopus