An Empirical Study on Retrieving Structural Clones Using Sequence Pattern Mining Algorithms
Yoshihisa Udagawa · 2014
Many clone detection techniques focus on fragments of duplicated code, i.e., simple clones. Structural clones are simple clones within a syntactic boundary that are good candidates for refactoring. In this paper, a new approach for detection of structural clones in source code is presented. The proposed approach is parse-tree-based and is enhanced by frequent subsequence mining. It comprises three stages: preprocessing, mining frequent statement sequences, and fine-matching for structural clones using a modified longest common subsequence (LCS) algorithm. The lengths of control statements in a programming language and method identifiers differ; thus, a conventional LCS algorithm does not return the expected length of matched identifiers. We propose an encoding algorithm for control statements and method identifiers. Retrieval experiments were conducted using the Java SWING source code. The results show that the proposed data mining algorithm detects clones comprising 51 extracted statements. Our modified LCS algorithm retrieves a number of structural clones with arbitrary statement gaps.