Scholarlink

<p>The collaboration between universities and enterprises synergistically foster the exchange of knowledge and technology. Research on university-industry collaboration (UIC) plays a pivotal role in promoting innovation, facilitating the transformation of scientific and technological achievements, and addressing “stuck neck” problems in technological advancement. This study proposes a topic-institution graph-based methodology for recommending UIC, applying graph analysis to extract multiple types of entities, cooperation information between institutions, semantic correlations between science, technology, and innovation (ST&amp;I) topics, and co-occurrence relationships between institutions and ST&amp;I topics. Firstly, the SciBERT model is applied to construct the science network and technology network, capturing the latent semantic correlations between paper keywords and patent keywords, respectively. Then, the Leiden community detection algorithm and multi-dimensional indicators are employed to identify promising ST&amp;I topics within these networks. Next, a topic-institution graph is constructed, incorporating rich structural and semantic information between multiple entities. Finally, the heterogeneous graph attention network (HAN) model is applied to predict collaboration opportunities between universities and enterprises. Additionally, the interpretability of ChatGPT-4o helps provide insightful explanations for the recommended collaborations and validate the accuracy of results. An empirical study on artificial intelligence domain demonstrates the feasibility and reliability of the proposed methodology.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Recommending university-industry collaboration: a topic-institution graph-based solution

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Lu Huang,
  • Xiaoli Cao,
  • Hang Ren,
  • Guangchao Wang,
  • Yani Wang

摘要

The collaboration between universities and enterprises synergistically foster the exchange of knowledge and technology. Research on university-industry collaboration (UIC) plays a pivotal role in promoting innovation, facilitating the transformation of scientific and technological achievements, and addressing “stuck neck” problems in technological advancement. This study proposes a topic-institution graph-based methodology for recommending UIC, applying graph analysis to extract multiple types of entities, cooperation information between institutions, semantic correlations between science, technology, and innovation (ST&I) topics, and co-occurrence relationships between institutions and ST&I topics. Firstly, the SciBERT model is applied to construct the science network and technology network, capturing the latent semantic correlations between paper keywords and patent keywords, respectively. Then, the Leiden community detection algorithm and multi-dimensional indicators are employed to identify promising ST&I topics within these networks. Next, a topic-institution graph is constructed, incorporating rich structural and semantic information between multiple entities. Finally, the heterogeneous graph attention network (HAN) model is applied to predict collaboration opportunities between universities and enterprises. Additionally, the interpretability of ChatGPT-4o helps provide insightful explanations for the recommended collaborations and validate the accuracy of results. An empirical study on artificial intelligence domain demonstrates the feasibility and reliability of the proposed methodology.