New Paper with collaborators: Activity Transition Graph Generation: How Far Are We?

Activity Transition Graphs (ATGs) map how an Android app moves between screens, and they underpin research on app comprehension, design recovery, and testing — all of which assume the graph is accurate. We participated in a collaboration led by Jiakun Liu in a new TOSEM paper, Activity Transition Graph Generation: How Far Are We? that checks that assumption for the first time.

The paper is based on the first ground-truth benchmark by manually identifying every transition in 98 Android apps, then used it to evaluate seven popular ATG tools. None is complete or fully correct: every tool both reports transitions that don’t exist and misses real ones, and different tools recover different subsets. Taking the union of tools improves recall but adds more false positives — trading one problem for another.

The paper also pins down five reasons dynamic tools miss transitions, notably state abstractions too coarse to tell distinct screens apart. The takeaway: much ATG-based research rests on an unmeasured foundation, and this benchmark finally lets tools be judged against the truth.

Link to the paper can be found here. You can download it here.