A Recommendation-Based Web Usage Mining Model for a University Community

In this paper a recommendation model based on web usage mining applied to the navigation data from a community college is proposed. The model is composed by an offline and an online module. In the offline module, the web sessions are preprocessed and represented in a vector space model using the frequency of the URLs, and after are grouped based on similarity measure using the Bisecting K-Means clustering algorithm. In the online module, each cluster found is represented by association rules, and the clusters are used for recommending web pages to the users. The article presents the procedures of selection and removal of entries in the web log, preprocessing of web sessions, and the strategies used for the web page recommendation. Supervised validation was applied to the model, selecting a group of web sessions as queries to the system, and asking users to answer a survey on the output.