Interpretable Academic Team Formation on Heterogeneous Information Networks: Constructive Heuristics with Explicit Organizational Affiliation

Assembling expert teams under strict skill-coverage and communication-distance constraints is a fundamental challenge in collaborative knowledge work. Existing learning-based approaches excel at probabilistic link prediction but cannot reliably enforce hard logical constraints or provide interpretable justifications. This study presents a constructive heuristic framework for team formation on Heterogeneous Information Networks (HINs), integrating authors, papers, departments, and organizations into a unified graph database. Seven algorithms exploit distinct structural features—topological proximity, co-authorship history, organizational affiliation, citation impact, and temporal recency—and guarantee constraint satisfaction by construction. Experiments on a subset of the AMiner DBLP dataset (≈625,000 nodes, 973,000 edges) covering 12,045 formation requests across 147 configurations show that algorithm choice is the dominant determinant of runtime, while skill frequency governs feasibility: success rates decline from 81.2% under abundant keywords to 14.9% in the long tail. Algorithms further form statistically distinct clusters in communication cost, and specialized heuristics operate in nearly disjoint solution spaces—supporting a toolbox approach over single-algorithm deployment. These results provide actionable selection guidance: proximity-based algorithms for communication-efficient teams; citation- or recency-aware algorithms when impact matters; cohesion-based algorithms when internal collaboration is the priority.