Enhancing Research Collaboration: A Text-Based Framework Using NLP for Optimal Partner Selection

Collaborating researchers tend to be more productive, but choosing the right partners is crucial. Traditional research collaboration recommendation systems rely on subject-based co-authorship, citation, and co-citation networks. This study introduces a framework that assesses research similarity primarily through the texts of authors' published research articles. By analyzing the abstracts of these publications, the proposed method leverages natural language processing (NLP) techniques to identify the best match between researchers' work. Instead of using author keywords and titles, which can be either too vague or too concise, abstracts are used because they effectively summarize the research objectives and contributions. This methodology was implemented using a small dataset consisting of 17,189 data entries gathered from the Scopus database. The study stands out because it focuses solely on text rather than network-based approaches, enabling the discovery of highly novel yet promising collaborators based on research similarity. To accurately represent the authors' research profiles, tf-idf and doc2vec vector embedding techniques were employed. Three human evaluators scored the machine-generated outcomes from 0 to 10 to measure the framework's reliability. The findings demonstrated that the trained model produced outcomes with exceptional precision, and doc2vec surpassed tf-idf in terms of accuracy.