gml-aec-knowledge-graph

A curated knowledge graph and wiki of 112 peer-reviewed papers on Graph Machine Learning (GML) and Graph Neural Networks (GNNs) applied across Architecture, Engineering and Construction (AEC), assembled via a PRISMA 2020 systematic review (search dates: February–March 2026). The dataset is keyed by manuscript first-appearance reference number (1..112) and comprises: • a master NetworkX MultiDiGraph (corpus_graph.json) with 1,570 typed edges across 9 connection modes — citation lineage, category cluster, domain cluster, methodology similarity, infrastructure dependency, temporal proximity, NetworkX modularity community, and two deep-PDF detected modes (shared GNN architecture / shared framework); • a 5-layer per-paper JSON schema (knowledge_base.json) covering bibliographic, categorical, methodological, relational, and temporal-quality metadata; • seven overlay files (manual + heuristic methodology extraction, DOI back-fill, lineage proposals, Connected Papers bibliometric verification, deep-PDF technical extraction, paper descriptions); • 144 Markdown wiki pages (one per paper, plus category/domain/lineage/methods cross-cuts); • 162 technical-extraction Markdown + CSV files for meta-analytic writing, with detected architectures (23 catalogues), frameworks (33), benchmarks (35), loss functions (10), and activations (9); • a BibTeX corpus (corpus.bib) of 110 active DOIs. Each field carries one of four provenance tags: curated (manually entered, no tag), manual extraction (teal), heuristic (yellow — regex catalogue match), bibliometric verification (blue — Connected Papers cross-reference). The reference numbering in this dataset matches the renumbered manuscript (where paper [1] is the first paper appearing in the body text, etc.); a refnum_mapping.json file documents the OLD↔NEW correspondence to the original Excel-order numbering. PDF source files of the 112 papers are NOT redistributed (copyright); bibliographic metadata is included in knowledge_base.json with DOI links.