抄録
This dataset is the Role-based Research Knowledge Graph (R-RKG), a structured knowledge graph of Japanese research projects funded by mission-oriented funding agencies and active during fiscal years 2020–2025. It covers 20,854 projects supported by 25 funding agencies and overseen by 8 government ministries, including Japan's four principal mission-oriented agencies (JST, AMED, NEDO, and NICT), with a total funding allocation exceeding 6.8 trillion yen.
The distinguishing feature of R-RKG is that each project description is decomposed into five semantic roles—Policy Objective, Purpose, Target, Method, and Evaluation—using a pipeline that combines a large language model for role extraction with multilingual embedding-based clustering. These roles capture, respectively, the program-level policy rationale for funding, the project-level research goal, the object of study, the technical approach, and the evaluation criteria. Each role is associated with one of 149 semantic clusters (30 Policy Objective, 30 Purpose, 50 Target, 30 Method, and 9 Evaluation clusters), enabling cross-agency and cross-program comparison of research investment along policy-to-research linkages.
The graph is modelled as a typed directed graph that integrates the funding structure (Agency–Program–Project) with the five-role semantic structure, and includes funding allocation amounts (in Japanese yen), project periods, implementing organizations, and provenance metadata. It is provided in RDF/Turtle and JSON-LD. Class names, properties, IRIs, agency names, and cluster labels are given in English (with official English agency names), while free-text fields such as project titles and the representative sentences of each role cluster retain their original Japanese with @ja language tags. The dataset supports research on science-of-science, research-policy analysis, scientometrics, and the construction of research intelligence infrastructure.