Browserbite: Accurate Cross-Browser Testing via Machine Learning Over Image Features Nataliia Semenenko*, Tõnis Saar** and Marlon Dumas* *{nataliia,marlon.dumas}@ut.ee, Institute of Computer Science, University of Tartu, Estonia **tonis.saar@stacc.ee, Browsrbite and STACC, Tallinn, Estonia Outline • • • • Introduction Visual cross-browser testing Machine learning model Results and future work Cross-browser visual testing Where’s that button? Internet Explorer 9 Internet Explorer 8 Goal • Develop method for cross-browser visual layout testing • Replace human labor in visual testing • Evaluate detected errors Methods • DOM (Document Object Model) based: Mogotest (www.mogotest.com), Browsera (www.browsera.com) • Image processing – non-invasive black box testing – Our current approach Web page Static image Cross-Browser Visual testing Web page visual segmentation • Image segmentation into regions of interest (ROI) • ROI comparison www.htcomp.ee ROI comparison • • • • Position Size Geometry Correlation VS ROI from WIN7 Chrome ROI from WIN7 IE8 Visual testing results Web page Static image Image segmentation (into ROIs) ROI comparison • Test set of 140 web pages from alexa.com • 98% recall • 66% precision Example of true positive Example of false positive ROI comparison + ML Web page Static image Image segmentation (into ROIs) ROI comparison Classification Machine learning • 140 most popular websites of Estonia according to www.alexa.com • 1200 potential incompatibilities • 40 subjects from 6 countries • Two classes :False positive vs True postive • Each ROI pair had 8 judgments • Inter-rater reliability 0,94 ROI features • • • • • • 10 histogram bins Correlation index Horizontal and vertical position Horizontal and vertical size Configuration index Mismatch Density Machine learning • Neural network • Three layers • 11 neurons in hidden layer • Five-fold cross-validation • Classification tree Results and Conclusions Measure Neural network 0.75 Classification tree 0.844 0.98 0.82 0.792 0.886 0.79 0.78 0.81 0.923 Plain Browserbite Mogotest Precision 0.66 Recall F-score 0.964 Results and conclusions Tool Mogotest CrossCheck [1] WebDiff [2] BB+ML Precision 75% 36% 96% 1. 2. 21% Choudhary, S.R., Prasad, M.R., and Orso, A. (2012). CrossCheck: Combining Crawling and Differencing to Better Detect Cross-browser Incompatibilities in Web Applications. (ICST), 2012 IEEE Fifth International Conference On, pp. 171–180. Choudhary, S.R., Versee, H., and Orso, A. (2010). WEBDIFF: Automated identification of cross-browser issues in web applications. (ICSM), pp. 1–10. Future work • Combination of image processing and DOM methods • Dynamic content suppression Thank You!