Abstract : Exploratory search systems help users learn or investigate a topic. The richness of the linked open data can be used to assist this task. We present a method that selects and ranks linked data resources that are semantically related to the user's interest. The objective is to focus the user's attention on a meaningful subset of highly informative resources. We extended spreading activation to typed graphs and coupled it with a graph sampling technique. The results selection and ranking is performed on-the-fly and doesn't require pre-processing. This allows addressing remote SPARQL endpoints. We describe first implementation on top of DBpedia. It is used by the Discovery Hub exploratory search system to select interesting resources, to support faceted browsing of the results, to provide explanations and to offer redirections to third-party services. Results of a user evaluation conclude the article.