Scaling Author Identity Disambiguation to the World of Code: A Methodology

2026-07-07

AI Summary

This work addresses the pervasive ambiguity and duplication in author identity strings within ultra-large-scale code commit datasets, where conventional approaches often produce oversized clusters containing irrelevant developers due to over-merging. The authors propose a collaborative disambiguation framework that integrates graph-cutting with edge-level classification: leveraging betweenness centrality for graph partitioning and training a high-precision edge classifier (AUC = 0.99) using GitHub’s “no-reply” email labels. Complementary mechanisms—including fingerprint-based anti-aliasing, node filtering gating, and attribute blocklisting—are incorporated to enhance precision. Evaluated on the World of Code dataset comprising approximately 107 million author strings, the method substantially mitigates over-merging, reducing the largest cluster size from 170,431 to under 7,000 and improving gold-standard recall from 0.44 to 0.70. It further outperforms existing privacy-preserving global disambiguation techniques on a separate set of 21 million distinct GitHub identities, achieving an unprecedented balance between precision and recall at billion-scale.

Abstract

We describe the methodology used to alias the free-text author/committer identities of the entire World of Code (WoC) collection (version V2604, ~107M distinct author strings over ~6B commits) into canonical persons, extending the fingerprint-based anti-aliasing of ALFAA and the 38M-identity resolution of Fry et al. by an order of magnitude. At this scale the central problem is over-merge, not missed merges: a few bridge identities (bots, role accounts, placeholder emails, multi-author commit fields) transitively weld unrelated clusters through the global union step into million-member "mega-clusters." We report the full experimental record (more than twenty experiments, including unsuccessful ones) behind the deployed design. Node-level gates (information score, project spread, degree) preserve recall but cannot dissolve the mega-cluster; per-value blocklists of high-quality-but-shared attributes are recall-safe but cannot break a redundant mesh; the working composition is a betweenness cut over the exact union graph plus a per-edge classifier trained on 2.6M labels mined from GitHub no-reply identifiers. That classifier, filtering dormant cross-project shingle groups and joined by GitHub's own account assertions, then recovers the recall the precision work had foregone. Against human-adjudicated pairs the per-edge model transfers at AUC 0.99; end to end the largest cluster falls from 170,431 (and a predecessor's 3.0M) to under 7,000 w hile gold recall rises from 0.44 to 0.70 at increasing precision, and on an independent 21M-alias GitHub ground truth the final map outscores its predecessors and the published state of the art among global, privacy-preserving resolvers. The record doubles as a catalog of scale lessons: structural cuts do not transfer to edge sets they never saw, and recall-only and precision-only benchmarks invert verdicts unless read together.

Problem

• author identity disambiguation
• over-merge
• mega-clusters
• code commits
• entity resolution

Innovation

• author identity disambiguation
• World of Code
• graph-based clustering
• edge classifier
• mega-cluster resolution