Technologies for GQM-Based Metrics Recommender Systems: A Systematic Literature Review

Mirko Farina,Giancarlo Succi,Artem Kruglov,Anna Gorb

doi:10.1109/access.2022.3152397

Mirko Farina, Giancarlo Succi + Show 2 more

Open Access

https://doi.org/10.1109/access.2022.3152397

Copy DOI

Abstract

Purpose: With this Systematic Literature Review (SLR), we aim to discover technologies to construct a Goal-Question-Metrics (GQM) based metrics recommender for software developers. Since such a system has not yet been described in the literature, we decided to analyse the technologies used in three main components of recommender systems - data sets, algorithms, and recommendations - independently. Methods: To achieve our goal we performed - following the best norms in our discipline - a systematic literature review (SLR). We first identified, through searches aptly performed, 422 potentially relevant papers, from which we selected - after applying inclusion and exclusion criteria - 30 papers, which we eventually included in our final log. Results: Systems with textual data set preprocess information in nearly the same way and the majority use similarity scores to create recommendations. Systems with GQM-based algorithms consist of questionnaires and require users to explicitly answer questions to produce suggestions. With respect to the recommendations of reviewed systems, they range from application programming interfaces (APIs) to requirements, but no system presently recommends metrics. Conclusion: In our SLR we: (a) identified a sequence of the most popular steps for preprocessing in recommender systems; (b) proposed an optimisation strategy for such steps; (c) found out that the most promising approach includes both ranking and classification; and (d) established that there are no recommendation systems developed to date to process metrics.

Highlights

G IVEN the software’s immaterial nature, researchers attempted to study its characteristics and influencing factors using software metrics, that is, numeric values attached to particular attributes of artifacts related to the software development process [1], [2]
There are many types of recommender systems with hybrid filtering used for various purposes [25], but in conducting this Systematic Literature Review (SLR) we found that they have not yet been used to automatically generate metrics based on the GQM model
RQ2: WHICH ALGORITHMS UNDERLINE RECOMMENDER SYSTEMS FOR SOFTWARE DEVELOPERS? Among all the algorithms used by the authors of the papers we reviewed in our SLR we can identify 5 major groups: (a) classification, (b) clustering, (c) ranking, (d) heuristics based approaches, and (e) questionnaires

Summary

Introduction

G IVEN the software’s immaterial nature, researchers attempted to study its characteristics and influencing factors using software metrics, that is, numeric values attached to particular attributes of artifacts related to the software development process [1], [2]. Among them one of the most relevant is the Goal-Question-Metrics (GQM), developed by Basili et al in the ’80s [11], [12]. In recent decades this approach has gained substantial popularity [6, p. 45]; the proper formulation of GQM is not a trivial task and requires substantial expertise. The purpose of this SLR is to investigate and find out which technologies can be used to develop a recommendation algorithm for software engineers, so as to automatically suggest appropriate metrics for GQM models.

Objectives

Results

Discussion

Conclusion