Case-Based Collective Classification

This is the first paper on textual case-based reasoning to employ collective classification, a methodology for simultaneously classifying related cases that has consistently attained higher accuracies than standard classification approaches when cases are related. Thus far, case-based classifiers have not been examined for their use in collective classification. We introduce Case-Based Collective Classification (CBCC) and report that it outperforms a traditional case-based classifier on three tasks. We also address issues of case representation and feature weight learning for CBCC. In particular, we describe a cross-validation approach for tuning feature weights and show that it increases CBCC accuracy on these tasks.