Birds of a Feather Surf Together: Using Clustering Methods to Improve Navigation Prediction from Internet Log Files

Many systems attempt to forecast user navigation in the Internet through the use of past behavior, preferences and environmental factors. Most of these models overlook the possibility that users may have many diverse sets of preferences. For example, the same person may search for information in different ways at night (when they are pursuing their hobbies and interests) as opposed to during the day (when they are at work). Thus, most users may well have different sets of preferences at different times of the day and behave differently in accordance with those preferences. In this paper, we present clustering methods for creating time dependent models to predict user navigation patterns; these methods allow us to segment log files into appropriate groups of navigation behaviour. The benefits of these methods over more established methods are highlighted. An empirical analysis is carried out on a sample of usage logs for Wireless Application Protocol (WAP) browsing as empirical support for the technique.