In France, seven DTUs (Digital Thematic Universities) allow open access to more than 24 000 OERs (Open Educational Resources). A DTU is a thematic repository of OERs, all validated by the academic community and indexed using SupLomFR (the French declination for higher education of the LOM standard). The available pedagogical resources are of various nature (case study, lessons, exercises, simulation, virtual experimentation, additional materials to lessons, pedagogical kit, serious game, self-assessment, etc.) and various formats (pdf, audio, video, interactive or multimedia document, 3D, ...). Each OER can be freely accessed from the DTU’s portal, at anymoment, by anybody, from everywhere. The main difficulty for a learner is to find the resources linked with his/her pedagogical objectives and his/her thematic background, when browsing the huge offer provided by a DTU. Facing the huge numbers of OERs and not familiar with the SupLOMfr indexing, most of users leave the DTU‘s portal without finding pertinent pedagogical materials. Thus it is important to assist the user by a recommender system that suggests pertinent and adequate resources to him/her. In addition, it is all the more important that is in open education. This paper describes a recommender system relying on the last resources the user has consulted: the recommender system takes into account the fact that a resource has been accessed as well as its description in SupLOMfr, if available. The interest of using information such as disciplines and keywords is to recommend the most adequate resources. Indeed, in the context of e-learning, it is crucial to make accurate predictions : a recommender with a low quality of prediction is not acceptable. The quality of prediction can be highly affected by the scarcity of resources: a problem appears when the last resource viewed by a user is an isolated resource (no similar resource in terms of keywords and disciplines exists).