Mapping scientific communities at scale

2025-01-17

AI Summary

This study addresses the dual challenges of scientific community structure analysis and emerging topic identification. Methodologically, it introduces the first end-to-end community mapping framework for large-scale scholarly corpora, integrating ScanR metadata, the OpenAlex citation network, and LLM-driven (Mistral Nemo) thematic annotation; cross-scale entity disambiguation is achieved via ORCID/ISNI/MeSH identifier harmonization, while strongly interactive filtering and multi-source fusion enable scalable research collaboration graph construction. Visualization—from national to laboratory level—is realized using Graphology (ForceAtlas2 + Louvain) and VOSviewer. Key contributions include: (1) a lightweight architecture supporting institutional-level embedded deployment; (2) a persistent-identifier–driven, high-precision disambiguation mechanism; and (3) real-time generation of high signal-to-noise-ratio collaboration graphs. The framework has been deployed in France’s national research network analytics platform, with its fully open-source toolchain serving over 100 research institutions.

Technology Category

• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Application Domains: Humanities & Computational Social Science
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Web Mining and Content Analysis: Community question answering

Abstract

This study introduces a novel methodology for mapping scientific communities at scale, addressing challenges associated with network analysis in large bibliometric datasets. By leveraging enriched publication metadata from the French research portal scanR and applying advanced filtering techniques to prioritize the strongest interactions between entities, we construct detailed, scalable network maps. These maps are enhanced through systematic disambiguation of authors, affiliations, and topics using persistent identifiers and specialized algorithms. The proposed framework integrates Elasticsearch for efficient data aggregation, Graphology for network spatialization (Force Atltas2) and community detection (Louvain algorithm) and VOSviewer for network vizualization. A Large Language Model (Mistral Nemo) is used to label the communities detected and OpenAlex data helps to enrich the results with citation counts estimation to detect hot topics. This scalable approach enables insightful exploration of research collaborations and thematic structures, with potential applications for strategic decision-making in science policy and funding. These web tools are effective at the global (national) scale but are also available (and can be integrated via iframes) on the perimeter of any French research institution (from large research organisms to any laboratory). The scanR community analysis tool is available online [https://scanr.enseignementsup-recherche.gouv.fr/networks/get-started](https://scanr.enseignementsup-recherche.gouv.fr/networks/get-started). All tools and methodologies are open-source on the repo [https://github.com/dataesr/scanr-ui](https://github.com/dataesr/scanr-ui)

Problem

• Scientific Community Analysis
• Visualization of Research Patterns
• Identification of Hot Topics

Innovation

• scanR Platform
• Scientific Collaboration Mapping
• Data Visualization Tools