OpenRank: A centrality algorithm for high-dimensional heterogeneous networks in open source collaboration

Heterogeneous information networks (HINs), composed of diverse node and edge types, provide a powerful paradigm for representing complex relational systems. In open-source collaboration, interactions among developers, repositories, and contributions naturally form high-dimensional heterogeneous structures. However, most existing centrality measures are designed for homogeneous or bipartite graphs and thus fail to capture the semantic diversity and temporal evolution inherent in such systems. To address these limitations, we propose OpenRank, a generalized centrality algorithm for heterogeneous networks that models inter-type influence via non-scalar, multi-type transfer matrices. OpenRank derives a closed-form iterative solution with provable convergence, offering both interpretability and scalability for large dynamic networks. The algorithm preserves relational semantics through multi-criteria aggregation of edge attributes and incorporates node-specific damping and domain-aligned initialization to reflect the evolving influence patterns across different roles and collaboration types. Extensive experiments on large-scale open-source collaboration datasets demonstrate that OpenRank achieves competitive adaptability to dynamic structural changes in dynamic propagation models and faster convergence compared with HEAT, HGT, and EdgeGFL. Comprehensive ablation and sensitivity analyses confirm the soundness and robustness of the algorithmic design. The method has been fully open-sourced through the OpenDigger project¹ and is already deployed in real-world industrial environments, including Alibaba and Ant Group.