STIFA: Crowdsourced Mobile Testing Report Selection Based on Text and Image Fusion Analysis

Crowdsourced mobile testing has been widely used due to its convenience and high efficiency [10]. Crowdsourced workers complete testing tasks and record results in test reports. However, the problem of duplicate reports has prevented the efficiency of crowd-sourced mobile testing from further improving. Existing crowd-sourced testing report analysis techniques usually leverage screenshots and text descriptions independently, but fail to recognize the link between these two types of information. In this paper, we present a crowdsourced mobile testing report selection tool, namely STIFA, to extract image and text feature information in reports and establish an image-text-fusion bug context. Based on text and image fusion analysis results, STIFA performs cluster analysis and report selection. To evaluate, we employed STIFA to analyze 150 reports from 2 apps. The results show that STIFA can extract, on average, 95.23% text feature information and 84.15% image feature information. Besides, STIFA reaches an accuracy of 87.64% in detecting duplicate reports. The demo can be found at https://youtu.be/Gw6ptqyQbQY.