Interdisciplinary scientific collaborator recommendation via interest diffusion and collaborative affinity

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Identifying promising collaborators for emerging interdisciplinary research remains a challenging task in scientific collaborator recommendation. Unlike general collaboration prediction, this task requires not only assessing whether a candidate scholar is structurally reachable, but also anticipating the new disciplinary directions into which a principal investigator may expand. Existing methods have mainly emphasized either network proximity or topical similarity, with limited attention to the joint modeling of collaboration feasibility, directional evolution, and fine-grained topical compatibility. To address this problem, we propose I-CAR, an interdisciplinary scientific collaborator recommendation model that integrates collaborative affinity, interest diffusion, and topical semantic matching. I-CAR first constructs a time-decayed collaboration network from historical project co-occurrence records to capture the structural basis of potential collaboration. It then projects the principal investigator’s personalized propagation distribution from the collaboration network into the disciplinary space to infer emerging research directions and estimate candidates’ interest compatibility with those directions. Finally, semantic representations of project keywords are used to measure fine-grained topical matching, and all features are fused through supervised learning to generate ranked collaborator recommendations. Experiments on completed projects funded by the National Natural Science Foundation of China show that I-CAR achieves better overall performance than structure-based baselines, content-based baselines, and heuristic fusion methods on Hit@K, Recall@K, and MRR@K. At K = 10, the supervised variant achieves a Hit@10 of 36.6% and a Recall@10 of 22.1%. These findings suggest that interdisciplinary collaborator recommendation is shaped by the joint effects of structural reachability, future research-direction expansion, and topical alignment.