Measuring effectiveness of Distributed Information Retrieval (DIR) is essential for research and development and for monitoring search quality in dynamic environment. Numerous works have been done to propose new search models in the context of peer-to-peer information retrieval systems (P2P-IR). In this article, we are considering another problem, which is the global ranking of a set of results' lists coming from a large set of IR systems. In this article we define a new method for automatic aggregation of results which mixes these categories by allowing each peer to construct knowledge about other peers' relevance model using a learning method (Formal Concept Analysis). The idea is that each peer constructs relationships between past queries, returned documents and contributed peers