An Algorithm for Conversational Case-Based Reasoning in Classification Tasks.

An important benefit of conversational case-based reasoning (CCBR) in applications such as customer help-desk support is the ability to solve problems by asking a small number of well-selected questions. However, there have been few investigations of the effectiveness of CCBR in classification problem solving, or its ability to compete with k-NN and other machine learning algorithms in terms of accuracy. We present a CCBR algorithm for classification tasks and demonstrate its ability to achieve high levels of problem-solving efficiency, while often equaling or exceeding the accuracy of k-NN and C4.5, a widely used algorithm for decision tree learning.