Students in a learning experience can be seen as a community working simultaneously (and in some cases collaboratively) in a set of activities. During these working sessions, students carry out numerous actions that affect their learning. But those actions happening outside a class or the Learning Management System cannot be easily observed. This paper presents a technique to widen the observability of these actions. The set of documents browsed by the students in a course was recorded during a period of eight weeks. These documents are then processed and the set with highest similarity with the course notes are selected and recommended back to all the students. The main problem is that this user community visits thousands of documents and only a small percent of them are suitable for recommendation. Using a combination of lexican analysis and information retrieval techniques, a fully automatic procedure to analyze these documents, classify them and select the most relevant ones is presented. The approach has been validated with an empirical study in an undergraduate engineering course with more than one hundred students. The recommended resources were rated as “relevant to the course” by the seven instructors with teaching duties in the course.