Advanced methods for Natural Language Processing and the availability of open knowledge sources, such as Wikipedia and BabelNet, have promoted recent progress in the field of content-based recommender systems (CBRSs). Those systems analyze both item descriptions (content) and user ratings to infer user profiles, which store information about preferences, exploited to suggest items similar to those users liked in the past. Novel research works have introduced Deep Content Analytics methods, i.e. semantic techniques that allow a better understanding of item properties, in terms of concepts instead of keywords. CBRSs can benefit from these techniques to implement more advanced, meaningful representations of items and user profiles. The talk will provide a basic survey of semantic techniques: top-down approaches, based on the use of different open knowledge sources (ontologies, Wikipedia, DBpedia, BabelNet) bottom-up approaches, based on the distributional hypothesis, which states that "words that occur in the same contexts tend to have similar meanings". The talk will show a semantic approach designed to provide explanations of suggestions, and a method for the discovery of hidden correlations among items, exploited to find “serendipitous” recommendations, i.e. items which are unexpected and potentially interesting at the same time.