Generalizing the Claim Lattice with Applications for Learning by Local Clustering

Learning Feature Relevance: Empirical and Forman Analyses

A central problem in case-based reasoning systems is that of salience: which features are important in determining similarity, and how might the closeness of situations be evaluated on the basis of their features. In this paper we use a graphical representation of a feature space that represents points in the space as nodes and links adjacent points by edges. This structure supports certain local reasoning within the feature space, allows the salience of features in particular situations to be learned over time, and provides a basis for predicting from situations of limited data.