Fig. 4: Identifying changes triggered by review comments
(Section V-B) of the comments in our oracle. To validate our tree-based model, we employed 10-fold cross-validation [26] and repeated the process 100 times. Finally, to validate our model via review participants, we sent five developers (three of them were not part of the interviews) a list of review comments they received for their recent code reviews and asked them to classify each comment as ‘Useful’ or ‘Not Useful’. We then compared these classifications with the classifications produced by the decision tree. The results of this validation are discussed below (Section V-F).