We propose an approach based on Linked Data for improving user satisfaction during her interactions with an information retrieval system. We want to provide the user with good search results snippets (i.e. snippets that let her quickly discover in which way a given document matches her information need). To attain this objective we use both the Web of data and the Web of documents. Firstly, for a given document we produce a sub-graph from the Web of data. Then, we use text analysis techniques (latent semantic analysis, co-clustering, ...) on textual resources associated with this graph in order to discover structures that can be used to enhance the initial graph.