Quantifying the Knowledge Proximity Between Academic and Industry Research: An Entity and Semantic Perspective

The academia and industry are characterized by a reciprocal shaping and dynamic feedback mechanism. Despite distinct institutional logics, they have adapted closely in collaborative publishing and talent mobility, demonstrating tension between institutional divergence and intensive collaboration. Existing studies on their knowledge proximity mainly rely on macro indicators such as the number of collaborative papers or patents, lacking an analysis of knowledge units in the literature. This has led to an insufficient grasp of fine-grained knowledge proximity between industry and academia, potentially undermining collaboration frameworks and resource allocation efficiency. To remedy the limitation, this study quantifies the trajectory of academia-industry co-evolution through fine-grained entities and semantic space. In the entity measurement part, we extract fine-grained knowledge entities via pre-trained models, measure sequence overlaps using cosine similarity, and analyze topological features through complex network analysis. At the semantic level, we employ unsupervised contrastive learning to quantify convergence in semantic spaces by measuring cross-institutional textual similarities. Finally, we use citation distribution patterns to examine correlations between bidirectional knowledge flows and similarity. Analysis reveals that knowledge proximity between academia and industry rises, particularly following technological change. This provides textual evidence of bidirectional adaptation in co-evolution. Additionally, academia’s knowledge dominance weakens during technological paradigm shifts. The dataset and code for this paper can be accessed at https://github.com/tinierZhao/Academic-Industrial-associations.

💡 Research Summary

This paper tackles the problem of measuring knowledge proximity between academia and industry at a fine‑grained level, moving beyond traditional macro indicators such as counts of joint papers or patents. The authors propose a two‑layer quantitative framework: an entity‑level analysis and a semantic‑space analysis.

At the entity level, a pre‑trained language model (BERT) is used to extract detailed knowledge entities—research methods, datasets, tools, evaluation criteria—from the full text of academic papers and industrial patents or reports in the NLP domain (2000‑2022). The extracted entity sequences for academia and industry are aligned and their overlap is quantified with cosine similarity. The entities are also linked through citation and co‑citation relations, forming a complex network whose topological properties (clustering coefficient, average path length, centrality measures) are examined.

The semantic layer employs unsupervised contrastive learning to embed the same documents into a dense vector space. Positive pairs (same topic across sectors) and negative pairs (different topics) are used to train the encoder, yielding cross‑institutional embeddings whose cosine similarity reflects semantic proximity.

Finally, citation distribution patterns are modeled over time, and statistical tests (Pearson correlation, Granger causality) are applied to assess the relationship between knowledge flow and the two proximity measures.

Empirical findings show that both entity overlap and semantic similarity increase markedly after major technological shifts—particularly the advent of deep learning and transformer models. During paradigm changes, academia’s dominance in unique entities declines while industry contributes a surge of new entities, indicating a rebalancing of knowledge leadership. Higher proximity correlates with stronger bidirectional citation flows; periods of heightened proximity exhibit a more than 30 % rise in reciprocal citations.

The study argues that fine‑grained entity and semantic analyses provide a more precise diagnostic tool for policy makers and managers seeking to foster effective academia‑industry collaboration. By releasing the dataset and code, the authors enable replication and extension to other fields.