It is critical to help researchers identify appropriate leaders to form teams to undertake new projects. Similarly, leading researchers also need to identify appropriate participating researchers in their research. To the research policymakers, identifying the research leadership can also facilitate fair and efficient fund allocations. Despite these pressing needs, research on leadership recommendation is yet to come. Specifically, the existing work has two main limitations. First, the focus of all the existing studies has been primarily on the recommendation of relationships among all co-authors while ignoring the role of leadership. It has been recognized that the research leadership can better capture the critical structure of research collaborations. Second, certain key proximities, such as geographical proximity and institutional proximity, have been found to be associated with the research collaboration and leadership flow. However, these proximities haven’t been fully used in the prediction tasks.

To address the above issues, we propose a novel model to capture the research leadership relationships and consider cognitive proximity, geographical proximity, and institutional proximity as node attribute information. The research leadership relationships are extracted from co-authored publications retrieved from the Web of Science (WOS) core citation database. The cognitive proximity denotes the cognitive similarity between researchers. We adopt natural language processing approaches to extract the cognitive background of researchers from their previous publications’ abstracts. Regarding geographical proximity, the closer the two researchers are, the larger the geographical proximity will be. Regarding institutional proximity, the more similar two researchers’ official language is, the larger their institutional proximity will be.

Taken together, we propose a hybrid model based on deep neural networks to preserve the above network topological features, and researchers’ attribute features, so as to conduct research leadership recommendations. Different from traditional research collaboration recommendations, the research leadership recommendation involves both link prediction and direction prediction. The link prediction aims to predict whether two authors will form a research leadership relationship in the future, and this relationship is undirected. The direction prediction task aims to predict who will become the research leader in the future. In sum, the main contributions of this work can be summarized as follows:

1. This research is the first machine-learning model to recommend research leadership in research collaboration by predicting the directed links in the research leadership network.
2. This research integrates a comprehensive set of proximities to facilitate the research leadership recommendation. Two critical yet under-utilized proximities, geographical and institutional proximities, are incorporated to facilitate the research leadership recommendation task.
3. This research represents a novel machine-learning framework that jointly considers the network features and node attribute information to make recommendations.

Extensive experiments and ablation studies have been conducted, demonstrating that our proposed model significantly outperforms the state-of-the-art collaborator recommendation models in research leadership recommendation. Our proposed model will lead to applications that facilitate both individual researchers and policymakers of institutions and funding agencies. On the one hand, for researchers with good research profiles but limited resources, our model can help them identify appropriate leaders to form joint efforts in research. On the other hand, for researchers with innovative ideas and resources, our model can help them identify participating researchers with suitable backgrounds to pursue the research. This is of particular importance to the formation of the multidisciplinary team to research emerging and cutting-edge research problems. By examining the dynamics of research leadership, individual researchers and policymakers can further characterize the evolution of research topics and trends in the field. For institutions, our model can help the administration team facilitate the growth of internal and external research leadership networks and provide a less biased measure of the research leadership of researchers. This is helpful in recruitment and promotion practice. For authorities and funding agencies, our model represents a data-driven measure of a researcher’s prospective leadership role in the field. Deserving researchers, particularly the rising stars, can be identified and allocated with more resources to nurture their growth.

The original article that this translation article is based on is: He, Chaocheng, Jiang Wu, and Qingpeng Zhang. “Proximity‐aware research leadership recommendation in research collaboration via deep neural networks.” Journal of the Association for Information Science and Technology (2021). DOI: https://doi.org/10.1002/asi.24546