Several algorithms of the top-k retrieval problem have been introduced in recent years. Unfortunately, current top-k query processing techniques focus on Boolean queries, and cannot be applied to the large Data Bases (DB) seen the gigantic number of data. In this paper, we propose an intelligent approach for top-k flexible queries taking into account another degree of granularity in the process of the evaluation of the query. It consists of two steps: 1) Generate a Meta-DB formed by a set of clusters resulting of a preliminary fuzzy clustering on the data. This set represents a reduced view of the initial DB and permits to deduct the semantics of the initial DB, and 2) Generate automatically the top k results using this Meta-DB and a new operator called stratified operator. We prove that this approach is optimum sight that the evaluation of the query is not done on the set of starting data which are enormous but rather by using the Meta-DB. Moreover, this approach can be applied to the large DB and that it does not require modifying the SQL language.