Explainable and Interactive Scientometrics with Large Language Models and Knowledge Graphs
Mirka Saarela, António Correia, Tommi Kärkkäinen · 2025
As scientometric data increasingly informs decision-making across academia, industry, and policy, existing analytical methods struggle to keep pace with the growing volume and contextual diversity of research outputs. While large language models (LLMs) and knowledge graphs (KGs) offer promising capabilities for tasks such as literature summarization, topic detection, and citation context analysis, their integration into scientometric workflows remains limited. This paper proposes a conceptual and technical roadmap for integrating traditional bibliometric techniques with LLM- and KG-based systems to deliver more contextualized, interactive, and explainable analyses. We introduce a system architecture that merges structured knowledge representations with generative AI, supported by reinforcement learning from human feedback, entity disambiguation workflows, and faceted search functionalities. This approach addresses persistent challenges in granularity, transparency, and adaptability of scientometric evaluations, while enhancing scientific knowledge representation, discovery, and assessment. By grounding LLM outputs in semantically rich KGs and incorporating human-centered AI practices, our framework advances toward more accountable, nuanced, and actionable scientometric applications. We outline implications for research evaluation, knowledge discovery, and interactive decision support, and discuss future directions for explainable and trustworthy AI in scientometrics.