A hybrid AI approach for recommending collaborators in research projects

Abstract

The success of research project proposals heavily depends on the consortium, which should be experienced and knowledgeable in the topics outlined in the corresponding calls, e.g., those in the EU’s research and innovation programme Horizon Europe. Yet, one of the most challenging activities in such a context is the formation of the consortium, which requires the identification of adequate research collaborators. Traditional methods take this challenge by relying solely on social networks and, or the number of author citations, which proved to be limited in efficacy. This paper proposes an Agentic Graph Retrieval-Augmented Generation (RAG) method, that provides contextual and explainable recommendations, which are tailored to researchers’ areas of expertise and project relevance, thus more effective than existing approaches. The proposed method combines Knowledge Graphs (KGs) and Large Language Models (LLMs) capabilities and has been developed following the Design Science research methodology. The new method has been evaluated by considering two of the highest-performant LLMs currently in the market: Claude Sonnet 3.5 and GPT-4o.