MetaGraph for FinNLP Research: A Large-Scale Knowledge Graph of GenAI in Financial NLP (2022–2025)

MetaGraph for FinNLP Research: A Large-Scale Knowledge Graph of GenAI in Financial NLP (2022–2025) This dataset accompanies the paper "Prompting the Market? A Large-Scale Meta-Analysis of GenAI in Financial NLP (2022–2025)". It introduces MetaGraph, a machine-readable knowledge graph automatically constructed from 681 scientific papers (see note below) in the field of financial NLP. MetaGraph enables large-scale meta-analyses of research trends, conceptual relationships, and method co-usage in GenAI for finance. Content Overview MetaGraph represents the FinNLP research landscape as a structured graph where: Nodes correspond to papers, models, datasets, tasks, and key research concepts. Edges encode citation links, shared methods, topic similarity, and temporal co-evolution. The graph facilitates: Discovery of underexplored intersections in FinNLP research Temporal analyses of GenAI adoption Visual exploration of conceptual and methodological trends Included Files finnlp_ontology.json: Defines an extensible ontology of FinNLP concepts (e.g., model, dataset, task) and relation types. finnlp_graph.graphml: The full MetaGraph in GraphML format, compatible with tools like Gephi, NetworkX, and GraphViz. finnlp_graph.json: A simplified JSON version for human-readable access, including structured metadata for nodes and edges. Use CasesThis dataset is intended for researchers in NLP, finance, and scientometrics. It can be used to: Conduct bibliometric studies Analyze the methodological evolution of GenAI in finance Train or evaluate systems for scientific knowledge extraction or reasoning CitationIf you use this resource, please cite the accompanying paper. Note: 58 of the 681 papers used in our analysis are not included in the published dataset. These papers were posted to ArXiv under a CC BY-NC-ND 4.0 license, which prevents reusers from distributing any derivative, adapted form of the original material. As such, we will not be distributing them as part of the MetaGraph knowledge graph.