This paper describes the participation of the Universities of Helsinki and Caen in the first round of the TREC Knowledge Base Acceleration track3. The task focused on filtering a stream of documents relevant to a set of entities. Our approach uses word co-occurrence graphs for modelling the named entities. We submitted two runs that achieved an average F-measure superior to the mean of all submitted runs. The best of those runs ranked in the top 5 runs for both the central and relevant F-measures, out of a total of 43 runs submitted by 11 institutions. As our runs were the produce of a first implementation of our approach these preliminary results are very supportive of our idea to use concept graphs for modelling named entity relations.