Abstract : Semantic web applications often access distributed triple stores relying on different ontologies and maintaining bases of RDF annotations about different domains. Use cases often involve queries which results combine pieces of annotations distributed over several bases maintained on different servers. In this context, one key issue is to characterize the content of RDF bases to be able to identify their potential contributions to the processing of a query. In this paper we propose an algorithm to extract a compact representation of the content of an RDF repository. We first improve the canonical representation of RDF graphs based on DFS code proposed in the literature. We then provide a join operator to significantly reduce the number of frequent graph patterns generated from the analysis of the content of the base, and we reduce the index size by keeping only the graph patterns with maximal coverage. Our algorithm has been tested on different data sets as discussed in conclusion.