On the Relation between the Context of aFeature and the Domain Theory in

Determining the context of a feature (i.e., the factors aaect-ing the ranking of a feature within a case) has been subject of several studies in analysis tasks, particularly for classiication but not in synthesis tasks like planning. In this paper we will address this problem and explain how the domain theory plays a key role in determining the context of a feature. We provide a characterization of the domain theory and show that in domains meeting this characterization, the context can be simpliied. We also use explanation-based learning techniques to determine the context in domains not meeting the characterization. Our work relates for the rst time CBR, machine learning and planning theory to determine the context of a feature.