Analysis of review quality by using gaze data during document review

In software development, deliverables in an upstream process are reviewed to ensure their quality and to reduce error propagation to the downstream process. Methods are available for evaluating the review quality. In this study, we considered the defect detection process in a review of Requirement Definition Documents and tested a potential relationship between the gaze patterns and review quality. Specifically, we analyzed the relationship between the gaze patterns, with a primary focus on the blink rate, in a review of RDDs and detection accuracy. A significant nonlinear correlation between the blink rate and the detection accuracy was observed; moreover, the subsequent regression analysis also verified the blink rate as the best predictor of the review quality, notwithstanding the use of other gaze patterns. This result indicates that the blink rate is a major predictor of a type of review performance.