Abstract : As part of our research is based on the extraction and analysis of knowledge from a documentary data source types that characterize legal counterfeit brand names. This discipline reflects perfectly all constraints belonging to the different areas involved in the context of knowledge extraction from documents: electronic documents, databases, statistics, artificial intelligence and man / machine interaction. However, the performance of these methods are closely related to the quality of the data used. In our research context, every decision is overseen by an editor (the magistrate) and depends heavily on the editorial context, limiting information extraction processes. So we are interested in decisions likely to bias learning materials. We see the foundations of them, determine their strategic importance and if necessary we offer tailored solutions to redirect the observed bias towards a better representation of documents. We offer a supervised exploratory approach to assess the quality of data involved, determining the properties skewing the quality of established knowledge and an interactive and collaborative platform modeling process leading to knowledge discovery in order to integrate effectively the expertise of the expert.