This article presents a recommendation-based web content mining model applied to the navigation data from an university community. The recommendation is based on an offline module that does the grouping of the web documents using a vector space model and the Bisecting KMeans algorithm, and an online module that selects the closest cluster and documents of the query document (actual navigation). Tasks for preprocessing the web documents, recommendation strategies, experiments, and a supervised validation are presented. The results suggests that the relationship between the query document and recommendations are good for near half of those polled.