Who Becomes a Key Developer? Relation-Aware Graph Learning on Artifact Traceability Graphs for Early Prediction

OverviewThe study investigates how developer–artifact relationships can improve the prediction of future key developers from their first six months of activity. The proposed approach constructs heterogeneous artifact traceability graphs representing developers, pull requests, issues, and files as nodes connected through typed relations and characterized by type-specific features.A customized Heterogeneous Attention Network (HAN) learns developer representations using developer-specific temporal masking. A hybrid meta-path selection procedure combines structural coverage and class discrimination to select complementary developer-to-developer interaction patterns of up to four hops.This package accompanies the experimental evaluation and analysis presented in the paper.Research QuestionsThe study addresses the following research questions:RQ1. To what extent does modeling developers’ initial activity using an artifact traceability graph improve the prediction of future key developers compared to the prior work?RQ2. Which developer–artifact interaction patterns are consistently selected and receive greater attention in predicting future key developers?