Abstract : The objective of the research performed here is to propose a methodology that guarantees a reduction in the number of criteria to be used for selecting partners (i.e. reduction of resource consumption) while allowing to choose the most representative criteria for users. The set of criteria chosen in this way does not influence the ranking of potential partners at all, or the influence is mastered and known by the user who has to interact with the algorithm. The methodology looks structuring criteria using the causal or semantic distance. To achieve this, the algorithms of "Dijkstra" and Ascending Hierarchical Clustering (AHC) have been used in sequence. This structuring is applied to a set of 101 criteria. The results are interpreted to demonstrate how the selection process can be optimized for a given usage scenario.