This work takes place in Knowledge Acquisition and Machine Learning research for incremental modelling and validation of problem solving knowledge. We introduce a simple model for representing procedural knowledge, based on the concept of situation, and we propose an incremental modelling and cooperative knowledge revision tool for Knowledge Bases (KB) expressed with that representation. This tool has been built and used for defining the sequence of French Minitel dialogues. In our situation-based model, each step of problem solving is explicitly expressed in the KBS as a simple and understandable object called "situation nodule". The corrections and completions of the KB are incrementally done, i.e. as new ill-solved cases are discovered, and in a cooperative way, i.e. relying on the user, who is the KB designer competent in the domain. The goals of our approach, which we suggest to call "situated revision", are the following: the revision process should be easy for the user, based on concrete cases, and carrying out "cautious" and validated corrections. The REVINOS tool has been developed with this perspective. Each revision step contains a KB objects modelling step, made by the designer, and then a semi-automatically correction step. REVINOS guides the designer during the whole revision process and suggests generalisations to similar concrete cases. One originality of REVINOS is to try to validate the repercussions of the suggested corrections, by submitting to the designer abstract examples that represent sets of concrete resolution cases.
