The effectiveness of a help desk system strongly depends on the ability to handle huge amounts of information. The precision of the information retrieval (IR) phase has a direct impact on both length and quality of the solution process and the improvement of precision is the focus of our research. Capitalizing on "type 2" fuzzy sets, we propose a keyword-based IR framework where relevance and confidence are explicitly modeled both in the system- user dialog and in the knowledge base management. Adaptivity features assure a constant evolution of the system knowledge and experiments prove that the proposed techniques actually improve the precision of the IR process. 12 Pages