Semantic social web becomes progressively real. Typed graphs are data structures constituted by a set of nodes connected by typed relations. Today, they are largely emerging due to semantic web achievements and may explode through growing synergies between social and semantic web. Their processing is difficult due to its large, heterogeneous and dynamic nature. At the same time their expressivity level and underlying schemas brought by semantics open new perspectives. In this general context, we address the specific use case of contextual and dynamic content recommendation in social networks. To achieve it we introduce the idea of a semantic spreading activation which uses graph semantics to increase its relevance. I will present corresponding mathematic model, proof of concept, early results and perspectives.