The Adaptive Webbrowser: Evaluation Step

Name	The Adaptive Webbrowser: Evaluation Step
Description	Compared to other methods such as neural networks and decision trees, the Bayesian classifier performes pretty well. In about 73% of the cases, the user agreed with the websites the adaptive browser suggested. A backpropagation neural network performed just as well, but decision trees performed worse. Additional features that can enhance the performance include predefining user profiles. Users themselves can provide words that they consider good indicators for websites of their interest. This way, the browser uses prior knowledge about the domain without having to learn it. Also, using lexical knowledge can enhance performance further. Using knowledge about relations between words can improve the quality of the list of most informative words. This lists always contains non-informative words like "other" and "however", because they don't appear in the stoplist (the list of words that can be skipped, like "what", "its" and "the"). Lexical knowledge can improve the quality of the list by removing some non-informative words.
Case Study	The Adaptive Webbrowser Case