ANN’s can be used in a similar way as Bayesian Networks to construct model
based RS’s. However, there is no conclusive study to whether ANN introduce any
performance gain. As a matter of fact, Pazzani and Billsus [57] did a comprehen
sive experimental study on the use of several machine learning algorithms for web
site recommendation. Their main goal was to compare the simple naive Bayesian
Classifier with computationally more expensive alternatives such as Decision Trees
and Neural Networks. Their experimental results show that Decision Trees perform
significantly worse. On the other hand ANN and the Bayesian classifier performed
similarly. They conclude that there does not seem to be a need for nonlinear classifiers such as the ANN. Berka et al. [31] used ANN to build an URL RS for web
navigation. They implemented a content independent system based exclusively on
trails – i.e. associating pairs of domain names with the number of people who traversed them. In order to do so they used feedforward Multilayer Perceptrons trained
with the Backpropagation algorithm.