Knowledge Graph-based Recommendation Systems for Academia-Industry Collaboration


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
  • Jian MA (Supervisor)
  • Xiuwu Liao (External person) (External Supervisor)
Award date8 Apr 2019


Academia-industry collaboration refers to the collaborative research, patent commercialization, contract research and consulting activities among academia and industry. It has shown to foster the knowledge transfer from academia to industry and realize the practical values of scientific knowledge. “Outline of the National Program for Long- and Medium-Term Scientific and Technological Development” states that technology innovation and science technology transfer are the objectives of our national science and technology system reform. “The Action Programme to Promote Scientific Achievements Transfer” and “The Construction Programme for National Technology Transfer System” issued by the State Council of China point out promoting academia-industry collaboration is the primary task of innovation-driven development strategy. Academia-industry collaboration is seen as a new engine to drive the economic development in the high-technology era. However, the academia-industry collaboration is ineffective as a large portion of research outputs are not utilized by industries. The gap between academic research and its contributions to the industry becomes increasingly larger. It is difficult for industries to access the research outputs and assess their practical values. This is mainly caused by the information asymmetry between the academia and industry. The industry is short of understanding for collaboration resources (mainly including academic patents and researchers) from the academia. 

Identifying academic patents and researchers is important for academia-industry collaboration. Effective recommendation systems could help the industry identify relevant resources for collaboration and promote academia-industry collaboration. However, existing patent and researcher recommendation systems cannot be directly applied to academia-industry collaboration context without considering of its characteristics. Effective patent/researcher recommendation systems are in need to ease the information asymmetry in academia-industry collaboration. Various entities are involved in academia-industry collaboration and relations between entities are diverse, for this reason, the recommendation systems for academia-industry collaboration should have the capacity to deal with heterogeneous information. Studies have shown that the knowledge graph has strength to organize heterogeneous information, and the rich semantic information contained in knowledge graphs could contribute to recommendation accuracy and interpretability. However, there lacks an applicable and general framework for knowledge graph-based recommendation. It needs further exploration to realize accurate and semantic recommendation by integrating the heterogeneous information with knowledge graphs. Therefore, this study proposes a knowledge graph-based recommendation framework and designs specific recommendation systems for academia-industry collaboration. The main contributions include:

(i) A novel knowledge graph-based recommendation framework is designed. In this framework, knowledge graphs are formally and generally defined, topological features based on relation paths are defined to analyze the semantic relatedness between entities, and several weight-learning strategies are proposed to learn recommendation models so as to combine multiple features for recommendation. The proposed recommendation framework bridges the research gap by providing an applicable and general framework for knowledge graph-based recommendation research and it can be flexibly applied to design specific recommendation systems by defining domain knowledge graphs and topological features.

(ii) A knowledge graph for academia-industry collaboration is designed. Based on the research of academia-industry collaboration, relevant entities (e.g., universities, researchers and companies) and their interactions (co-inventing patents and cooperating on R&D projects) are identified to define the schema of knowledge graph for academia-industry collaboration. A knowledge graph is constructed based on the defined schema. The designed knowledge graph is customized for academia-industry collaboration and bridges the gap that existing knowledge graphs (e.g., Google Knowledge Graph) are too general for applications in academia-industry collaboration domain. The designed knowledge graph lays a foundation for applications of academia-industry collaboration, and it could be utilized to design various recommendation systems for academia-industry collaboration.

(iii) A knowledge graph-based patent recommendation system is proposed to promote academic patent commercialization. Knowledge graph is utilized to organize information about academic patent commercialization. Topological features for patent recommendation are defined to capture the semantic relatedness between patent buyers and academic patents. Experiments are conducted based on the real data of academic patent commercialization. The knowledge graph-based patent recommendation system is shown to be more effective for academic patent commercialization than traditional recommendation models (e.g., collaborative filtering models, content-based models and citation network-based models). Knowledge graph-based topological features are also shown to be more effective at analyzing the semantic relatedness between entities. The proposed recommendation system could proactively generate accurate recommendations for the industry to promote academic patent commercialization.

(iv) A context-aware researcher recommendation system is proposed for R&D project collaboration. Knowledge graph for project collaboration is designed for context-aware researcher recommendation. Multiple recommendation features are defined to analyze researchers according to the important criteria for partner selection. A new feature metric is proposed to analyze the semantic relatedness between entities with the consideration of recommendation context. Experiment results have shown that the knowledge graph-based researcher recommendation system is more effective at identifying suitable researchers for R&D project collaboration than traditional expert recommendation methods (e.g., content-based and social network-based methods). The proposed feature metric is proved to be effective in the context-aware researcher recommendation. The proposed recommendation method could help companies identify suitable partners and facilitate academia-industry collaboration on R&D projects.

    Research areas

  • Knowledge graph, Recommendation systems, Academia-industry collaboration, Patent commercialization, Project collaboration