Methods for selecting and improving software clustering algorithms

Abstract

SUMMARYSeveral software clustering algorithms have been proposed in the literature, each with its own strengths and weaknesses. Most of these algorithms have been applied to particular software systems with considerable success. However, no algorithm has been shown to be superior in all cases. As a result, selecting a software clustering algorithm that is best suited for a specific software system remains a hard question to answer. At the same time, improving the effectiveness of an existing algorithm is a time‐consuming process that would benefit from a methodology that allowed the early evaluation of an idea. In this paper, we first introduce a formal description template for software clustering algorithms. Based on this template, we propose a novel method for the selection of a software clustering algorithm for specific needs, as well as a method for software clustering algorithm improvement. The applicability and usefulness of the two methods introduced in this paper is demonstrated by applying them in four distinct case studies. Copyright © 2012 John Wiley & Sons, Ltd.

Improving software modularization via automated analysis of latent topics and dependencies

“…Anquetil and Lethbridge [Anquetil and Lethbridge 1999] tested some of the algorithms proposed by Wiggerts and compared their strengths and weaknesses when applied to system re-modularization. A more recent work by Shtern and Tzerpos [Shtern and Tzerpos 2009] introduced a method for selecting a clustering algorithm for the system decomposition given specifics needs. Wu et al [Wu et al 2005] describe a comparative study of clustering algorithms in the context of software evolution.…”

Cooperative based software clustering on dependency graphs

“…Evaluation methods involve an appraisal of software clustering approaches based on an assessment of the quality of an automatic decomposition using an authoritative decomposition as a reference. The drawback of these methods is the assumption that such a decomposition exists, given that its construction for a mid-sized software system is quite challenging, even with the help of an expert [55]. Numerous research studies have addressed the creation of an authoritative decomposition [31,43].…”

“…In the context of software clustering, Shten [55] suggests that software cluster analysis can be done in the following stages: (1) fact extraction, (2) ltering, (3) similarity computation, (4) cluster creation, (5) results visualization, and (6) user feedback collection. The process typically repeats until satisfactory results obtained.…”

“…Some software clustering algorithms accept (or require) user input, whereas others run totally automatically [55]. Algorithm which accept user input are called semiautomatic clustering algorithms.…”

Split-Jaccard Distance of Hierarchical Decompositions for Software Architecture

“…Software clustering techniques are commonly used for recovering architecture information such as a module view of the system. Various approaches to software clustering have been proposed, and each of them has strong and weak points from different aspects [1], [2].…”