How Does the Application Help Researchers Connect?

The application analyzes scientific publications and generates a research profile based on them. It then visualizes connections between researchers, research themes, and collaboration networks. This provides insight into who is working on which topics, which researchers are already collaborating, and where potential new connections may exist. As such, the tool serves as a valuable resource for finding new research partners within VU.   

To categorize research interests, we use the United Nations’ Sustainable Development Goals (SDGs). These goals provide a broad framework for societal challenges and encompass not only environmental and sustainability issues, but also topics such as health, education, inequality, and economic development. As a result, the SDGs are relevant to a wide range of researchers, even when their work is not directly focused on sustainability.   

In addition to SDG classification, the application also displays substantive research themes derived from publications. These themes may be strongly, partially, or only indirectly related to sustainability and provide an additional perspective on a researcher’s expertise.   

By mapping individual profiles in this way, researchers with shared interests or complementary expertise can be easily identified. When two researchers have overlapping themes in their profiles, this creates opportunities for knowledge exchange, networking, and potential collaboration.  

How Are the Profiles Created?

For each abstract, we used GPT-4o-mini to estimate the extent to which the described research contributes to one or more Sustainable Development Goals (SDGs). In the future, we aim to train our own local model for this purpose, reducing our reliance on external services. Although we already have experience training models to classify policy documents, research publications require a different approach. The language, structure, and content of scientific texts differ significantly from policy documents, making additional model development necessary.   

In addition to SDG classification, we use the fingerprints and research topics assigned by Elsevier through Scopus. These topics provide insight into the research areas in which a researcher is active. Because some topics are very broad while others are highly specific, they are not always suitable as a concise summary of someone’s expertise. Therefore, we used GPT-4o-mini to translate these topics into broader research fields that are more useful for profile building and identifying similarities between researchers.   

We also applied BERTopic to all publications by VU researchers to identify underlying research themes. BERTopic uses embeddings, which are numerical representations of text that capture meaning and context. Based on these representations, the model groups words and documents that are semantically related into themes. This creates an overview of the main topics present within the publications.    

Each publication is linked to one dominant topic. By aggregating this information at the level of individual researchers, a profile is created that highlights the research themes most characteristic of their work.   

At the same time, we are developing more advanced network analyses to further refine research profiles. These analyses consider not only publication content, but also collaboration patterns based on thematic connections between researchers and shared co-authors. In this way, we aim to create an even richer picture of expertise and potential collaboration opportunities within VU.