In this work, the dynamics of interest topics in web-based educational environments are studied by analysing the students' interactions during the teaching/learning process. Forums are the main tools to observe these interactions for distance learning. Specific techniques are needed for analysing the large amount of data generated in the course, which allows faculty making interesting conclusions to improve it. In particular, we propose a set of algorithms to discover relevant topics (and subtopics) appeared in the forums' messages. The results obtained are also discussed in order to find out patterns of behaviour.