In one embodiment, a user interface includes at least one instance of each of at least one widget. Recording a plurality of widget interaction instances (WIIs) for the user interface, each WII resulting from a user interaction applied to a particular instance of a particular widget. Clustering the plurality of WIIs based on a text value and a path value of each WII, such that each cluster of WIIs is associated with a particular widget. Determining, for each of at least one cluster of WIIs, whether the particular widget associated with the cluster of WIIs is erroneous based on whether user interactions corresponding to the WIIs in the cluster have produced responses from a software application that includes the user interface.