Feature selection for clustering based aspect mining

説明

This paper proposes a new heuristic algorithm for optimizing the set of features of clustering based aspect mining that aims at identifying code which is likely to implement a crosscutting concern. Given a set of features, our algorithm selects important ones for clustering by using self-organizing maps (SOM). We implemented the algorithm by using the SOM Toolbox and evaluated its impact by evaluating the accuracy of aspect mining based on the optimized set of features. The results of experiments revealed that different programs have different optimal features and showed following improvements: 1) the accuracy of clustering concerns are increased even the number of features are decreased. 2) our algorithm successfully find the optimal set of features automatically against different programs.

収録刊行物

詳細情報 詳細情報について

問題の指摘

ページトップへ