Bridging Information Asymmetry in University–Industry Collaboration: MMKG SNR

Effective university–industry collaboration remains elusive, largely because firms, especially small and medium-sized enterprises, struggle to identify trustworthy academics whose expertise aligns with their technical needs. We address this problem by introducing MMKG-SNR, a ranking framework that fuses (i) a Multimodal Knowledge Graph (MMKG) grounding textual and visual evidence from publications and patents with (ii) an enriched Social-Network (SN) graph that captures citation ties, institutional links, and fine-grained company–scholar interaction logs obtained from the Jiangxi Online Technology Trading Platform (OTTP). Requirement texts are used as external keys in a hand-in-hand dual-stream Transformer to retrieve context from the MMKG, while RotatE embeddings model relational trust in the SN graph, augmented with newly formalized company–scholar edges. The two signal sources are combined through a query-conditioned gating mechanism, and a pairwise Bayesian Personalized Ranking loss trains a two-layer neural scorer to optimize the final ordering. Experiments on real OTTP collaboration cases show that MMKG-SNR improves recall and nDCG over strong content-only and trust-only baselines, demonstrating its ability to surface scholars who are both technically relevant and socially credible. Ablation analyses confirm that requirement-anchored keys, company–scholar edges, and the learn-to-rank objective each contribute significant, complementary gains. The framework provides a reproducible, extensible blueprint for reducing information asymmetry and accelerating knowledge transfer in technology-driven economies.