In the context of the Social Web, user’ profiles reflecting an individual’s interests are being modeled using semantic techniques that consider the users posts’ and take advantage of the rich background knowledge in a Linked Open Dataset (LOD). To enrich the user profile, expansion strategies are applied. While these strategies are useful in Social Network posts, their suitability for modeling users’ interests with larger documents as input has not yet been validated. Thus, we built a profile of user’s research interests to recommend academic documents of possible interest. Contrary to the results obtained in the Social Web, the expansion techniques are inadequate for the academic texts scenario when all of text in the documents are used as input. Our results show a new filtering strategy performs better in such a scenario. An additional contribution was our creation of a DBpedia annotated dataset for academic document recommendation, which was built from a corpus of open access papers available through Core and Arxiv. Findings suggest the need to further explore new strategies to construct semantic models that are able to operate in different domains.