Incremental Multistrategy Relational Conceptual Clustering and Ordering Effects

概要

論文の詳細を見る
We have developed an algorithm, called MMD, for the unsupervised discovery of error classes from relational descriptions of behavioral discrepancies via multistrategy conceptual clustering. The algorithm is incremental, and like all heuristic incremental algorithms that summarize information, it can suffer from so-called ordering effects : different orderings of the input data may yield different classifications. In this paper, we evaluate the ability of the causality component of MMD, and two other mechanisms, namely, general reclassification and feature weighting, to mitigate ordering effects. Experimental results show that MMD can effectively mitigate ordering effects while still achieving high performance in terms of discovering and detecting errors in novice programs.
社団法人人工知能学会の論文
1999-01-01