Looking for novelty in SBSE problems

Abstract

Search-based software engineering (SBSE) was conceived to support engineers searching for innovative ideas to solve recurrent software engineering problems along the software project lifecycle. However, current approaches require the engineer to formulate and quantify their search objectives, which may be challenging. As SBSE consolidates as a discipline, problems become more demanding, and consequently the definition of the search problem and the characteristics of the search space remain oversimplified. Thus the evaluation of problem solutions by means of a fitness function could be failing to address essential aspects that can cause disappointment for the engineer after reaching final results. This position paper launches the idea that novelty search opens up a new scenario, as it rewards solution novelty, a concept mapping to problem characteristics other than fitness and whose definition might be more intuitive to the engineer. We explore its applicability to SBSE and discuss some preliminary findings of interest to the SBSE community.