This paper is about the adaptation knowledge (\ak) discovery for the \taaable system, a case-based reasoning system that adapts cooking recipes to user constraints. The \ak comes from the interpretation of closed itemsets (\cis) whose items correspond to the ingredients that have to be removed, kept, or added. An original approach is proposed for building the context on which \ci extraction is performed. This approach focuses on a restrictive selection of objects and on a specific ranking based on the form of the \cis. Several experimentations are proposed in order to improve the quality of the \ak being extracted and to decrease the computation time. This chain of experiments can be seen as an iterative knowledge discovery process: the analysis following each experiment leads to a more sophisticated experiment until some concrete and useful results are obtained.