There are a lot of educational resources publicly available online. Recommender systems and information retrieval engines can help learners and educators navigate in these resources. However, the available educational resources differ in format, size, type, topics, etc. These differences complicate their use and manipulation which raised the need for having a common representation for educational resources and texts in general. Efforts have been made by the research community to create various techniques to homogeneously represent these resources. Although these representations have achieved incredible results in many tasks, they seem to be dependent on the writing style and not only on the content. Furthermore, they do not generate representations that reflect a semantic representation of the content. In this work, we present a new task-agnostic method (EMBEDD-ER) to generate representations for educational resources based on document annotation and Linked Open Data (LOD). It creates representations that are focused on the content, compact, and can be generalized to unseen resources without requiring extra training. The resulting representations encapsulate the information found in the resources and project similar resources closer to one another than to non-similar ones. Empirical tests have shown promising results both visually and in a subject classification task.