Evaluated on open-source Python repositories, the consensus fusion method enhances co-change prediction accuracy, indicating improved integration processes.
Predicting which files must co-change in a software commit is critical for reducing incomplete pull requests and integration failures. We introduce a consensus fusion approach that combines Katz centrality on import graphs (structural signal) with co-change frequency from version history (behavioral signal) using CombMNZ, a rank fusion method from information retrieval. Evaluated on 15 open-source Python repositories (688 commits), CombMNZ achieves +57% F1 (p=0.0001) over structural analysis alone and +16% F1 (p=0.012) over the ROSE association rule mining baseline. We prove that ROSE confidence-based ranking is mathematically equivalent to raw co-change frequency under single-seed evaluation, and show that 70.9% of correct predictions come exclusively from structural analysis while 21.6% come from history alone — confirming true signal complementarity. Our work transfers the chorus effect from information retrieval to software engineering.
No takes yet. Share an insight, caveat, or question.
Matthew Hume (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: