In this paper, we describe a novel framework for improving information retrieval results. At first, relevant documents are organized in clusters utilizing the containment metric along with language modeling tools. Then the final ranked list (ascending/descending order) of the documents that will be returned to the user for the specific query, is produced. To achieve that, firstly we extract the scores between the clusters and the query representations and then we combine the internal rankings of the documents inside the clusters using these scores as weighting factor. The method employed is based in the exploitation of the inter-documents similarities (lexical and/or semantics) after a sophisticated preprocessing. The experimental evaluation demonstrates that the proposed algorithm has the potential to improve the quality of the retrieved results.