BuildGraph: A Synthetic Multi-Archetype Building Knowledge Graph Dataset

BuildGraph is a dataset of 120 commercial building knowledge graphs in the Brick Schema ontology (v1.3), provided as RDF Turtle (.ttl) files. Buildings are parameterized from U.S. DOE Commercial Prototype Building Models across 8 building types (small office, medium office, large office, primary school, hospital, hotel, strip mall, outpatient healthcare) and 3 ASHRAE 90.1 energy code vintages (pre-1980, 2004, 2013), yielding 24 archetypes × 5 random seeds = 120 buildings.

Sensor placement probabilities are calibrated against 1,421 patterns mined from 65 real commercial buildings on the Mortar platform, included in this deposit as empirical_patterns.json. Each building independently samples a sensor-attachment style (brick:hasPoint / brick:isPointOf / both) and a URI naming convention (numeric, hyphenated, descriptive, or legacy), mirroring heterogeneity found in real deployed Brick buildings. Dataset scale ranges from ~100 triples (sparse pre-1980 strip mall) to ~11,000 triples (dense 2013 hotel), with a mean of 1,865 triples and 470 typed entities per building.

This deposit also includes sparql_queries.json, a benchmark of 75 SPARQL queries across 5 categories (equipment discovery, sensor/point relations, topological traversal, aggregation, multi-hop/complex) with automatic scope detection. BuildGraph buildings achieve an overall Query Answerability Rate (QAR) of 91.5% on this benchmark, compared to 35.8% for 59 publicly available real-world Brick building files. The generator code and benchmark scripts are available at https://github.com/humanbuildingsynergy/BuildGraph.