Source Code Retrieval Using Sequence Based Similarity

TL;DR

This paper presents a novel source code retrieval method that leverages structural information and a sequence-based similarity measure to effectively identify duplicated code in Java programs.

Contribution

It introduces a new similarity measure and retrieval algorithm that significantly improves code search accuracy by considering control statements and method identifiers.

Findings

Outperforms existing models by up to 90.9% in retrieval accuracy

Uses a sequence-based similarity measure extending set similarity indices

Develops a lexical parser for extracting structural code features

Abstract

Duplicated code has a negative impact on the quality of software systems and should be detected at least. In this paper, we discuss an approach that improves source code retrieval using the structural information about the programs. We developed a lexical parser to extract control statements and method identifiers from Java programs. We propose a similarity measure that is defined by the ratio of the number of sequentially full matching statements to the number of sequentially partial matching ones. The similarity measure is considered to be an extension of a set based similarity index, e.g., Sorensen-Dice index. Our key contribution of this research is the development of a similarity retrieval algorithm that derives meaningful search conditions from a given sequence, and then performs retrieval using all of the derived conditions. Experiments show that our retrieval model outperforms the other retrieval models up to 90.9% in the number of retrieved methods.