Two important notions are used: the neighborhood that allows the valuation of terms and the path of tags followed to reach lexical units. This similarity measure has been exploited by a system of semi-structured information retrieval which we realized.