Attribute completion developer recommendation algorithm based on decoupled graph convolution
LU Yan
LING Songsong
ZHONG Jing
YU Xu
Abstract：The crowdsourced software development model harnesses global developer resources to enable efficient software development,yet information overload on platforms poses challenges in recommending suitable developers.Existing recommendation algorithms rely on developer attributes and interaction be-haviors,but incomplete developer profiles in crowdsourced platforms hinder recommendation perform-ance.To address this,we propose an Attribute-Completion Developer Recommendation algorithm via De-coupled Graph Convolution(ADD).ADD comprises two modules:an attribute inference module and an attribute completion module.The attribute inference module leverages decoupled graph convolution to de-duce developer attributes across multiple latent spaces,while the attribute completion module fills missing attributes using inferred results.The completed attributes are iteratively fed back into the inference mod-ule,forming a cyclic optimization process until convergence.Upon convergence,the model achieves peak attribute inference capability,generating attributes that closely reflect real-world scenarios.Experiments on real-world datasets demonstrate that ADD significantly outperforms state-of-the-art models in MAE,RMSE,Precision,and Recall metrics,effectively resolving information incompleteness in crowdsourced developer recommendations.
Keywords：developer recommendationsdisentangled representation learningattribute completioncrowdsourced software development
Publication Date：2025-08-01
Online Publishing Date：2026-08-28(First online date of this platform, not the publication date of the document)
Pages：12( 485-496 )