Data Mining in Programs

Abstract

Software exists in various control systems, such as security-critical systems and so on. Existing program clustering methods are limited in identifying functional equivalent programs with different syntactic representations. To solve this problem, firstly, a clustering method based on structured metric vectors was proposed to quickly identify structurally similar programs from a large number of existing programs. Next, a clustering method based on similar execution value sequences was proposed, to accurately identify the functional equivalent programs with code variations. This approach has been applied in automatic program repair, to identify sample programs from a large pool of template programs. The average purity value is 0.95576 and the average entropy is 0.15497. This means that the clustering partition is consistent with the expected partition.