Multi-dimensional Assessment of Crowdsourced Testing Reports via LLMs

Crowdsourced testing can markedly enhance test coverage and the discovery rate of potential defects compared to traditional software testing, making it increasingly popular. However, with the widespread use of crowdsourced testing, more and more crowdworkers from various backgrounds are submitting a large number of testing reports to crowdsourced testing platforms, which hinders developers from effectively reviewing the reports. Facing a vast amount of reports with varying quality, manual review is not only time-consuming and labor-intensive but also increases costs. Therefore, how to efficiently review crowdsourced testing reports has become a major challenge. To address this challenge, we propose a multi-dimensional assessment method for crowdsourced testing reports based on large language models. This method not only inherits the textuality dimension widely used in traditional report assessment but also innovatively introduces two new dimensions: adequacy and competitiveness. It comprehensively assesses the quality of crowdsourced testing reports from multiple perspectives, aiming to better screen for high-quality crowdsourced testing reports. Through experimental analysis conducted on three different applications, we have proven the consistency of our method with human raters across various dimensions, and we have also observed an enhancement in the efficiency of report assessment.