Nature Research Intelligence Topics

Scientific research grows at a tremendous pace, and by classifying over 140 million documents (research articles, book chapters, pre-prints and more) into over 29,000 topics we help make research more discoverable by enabling tracking broad trends as well as supporting deep dives in specialised areas.We are releasing a dataset containing version 1 of the Nature Research Intelligence topics. The topics were identified by using machine learning to cluster and organise a large citation network, built from documents citing each other. The resulting clusters were labelled using generative AI. The full methodology is described in a pre-print: Jenset, Bevan & Jain (2025).The dataset has one row for each of the 29,140 topics (at the most granular level), as well as the header row. The columns in the file are as follows:topic_label: the label for the topic, created with generative AI based on documents in the topic.size: the number of documents in the topic, as of May 19, 2025.topic_coherence: a metric from 0 (no coherence) to 1 (max coherence) indicating how coherent the topic is.topic_if: an impact factor type of metric indicating average citations to documents in the topic, calculated using the standard formula.for_l3_parent, for_l2_parent, for_l1_parent: a hierarchy organising the topics into progressively broader fields, using the ANZSRC fields of research. For further details see:Gard B. Jenset, Peter J. Bevan, Akarsh Jain et al. A large-scale, granular topic classification system for scientific documents, 27 April 2025, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs-6529718/v1] [link]