Graph-mediated LLM exploration for IFC question answering : a benchmark comparison against direct-IFC approaches: Master’s Thesis

Bhatti, Haris Waheed (avtor), Turk, Žiga (mentor)

Industry Foundation Classes (IFC), the standard format for exchanging building information, are hard to query without specialist tooling, so most people working in construction cannot get information out of their own building models. This thesis asks whether mediating IFC through a knowledge graph and exploring it adaptively can compete with exploring raw IFC directly. We built ifctoneo4j, which converts IFC into a Neo4j property graph following the Linked Building Data ontologies, and BIMGraphExplorer, which lets an LLM query that graph in Cypher through Model Context Protocol tools, and evaluated both under matched conditions: same model, same judge, 367 questions across 23 IFC-Bench v2 datasets. The direct-IFC baseline wins by 10.3 percentage points (82.3 % versus 72.0 %) with equal transparency, though the graphmediated system wins outright on five of the twenty-three datasets, by as much as 24 points on one; most of the remaining gap comes from fixable problems in how the graph was built, with the one real ceiling on quantities Cypher cannot compute. This matters because building data can live in an open, shared graph format and still answer questions nearly as well as reading a proprietary file directly, which is useful for anyone building natural-language tools over construction data or working in Linked Building Data. What is left is mostly engineering work: improving how files are converted, and giving the system a way to compute a value instead of only looking one up, so it can also handle geometry and material questions the graph alone cannot answer.

• Jezik: Angleški jezik
• Ključne besede: master thesis, civil engineering, Industry Foundation Classes (IFC), Building Information Modelling (BIM), knowledge graph, Linked Building Data, large language models, adaptive exploration, natural-language question answering
• Vrsta gradiva: Magistrsko delo/naloga
• Tipologija: 2.09 - Magistrsko delo
• Organizacija: FGG - Fakulteta za gradbeništvo in geodezijo
• Kraj izida: Ljubljana
• Založnik: [H. W. Bhatti]
• Leto izida: 2026
• Št. strani: 1 spletni vir (1 datoteka PDF (IX, 54 str.))
• PID: 20.500.12556/RUL-185169
• UDK: 004.42:69(043.2)
• COBISS.SI-ID: 286173955
• Datum objave v RUL: 25.07.2026

Sekundarni jezik

• Jezik: Slovenski jezik

Izvleček

Datoteke Industry Foundation Classes (IFC), standardni format za izmenjavo informacij o stavbah, je brez specializiranih orodij težko poizvedovati, zato večina ljudi, ki dela v gradbeništvu, ne more pridobiti informacij iz lastnih modelov stavb. Magistrsko delo preverja, ali lahko posredovanje IFC preko znanjskega grafa in njegovo prilagodljivo raziskovanje tekmuje z neposrednim raziskovanjem surovega IFC. Zgradili smo ifctoneo4j, ki pretvori IFC v lastnostni graf Neo4j po ontologijah Linked Building Data, in BIMGraphExplorer, ki jezikovnemu modelu omogoča poizvedovanje po tem grafu v jeziku Cypher prek orodij protokola Model Context Protocol, oba pa smo ovrednotili pod enakimi pogoji: isti model, isti ocenjevalec, 367 vprašanj na 23 naborih podatkov IFC-Bench v2. Neposredni pristop IFC zmaga za 10,3 odstotne točke (82,3 % proti 72,0 %) ob enaki transparentnosti, čeprav sistem s posredovanjem preko grafa v celoti zmaga na petih od triindvajsetih naborov podatkov, na enem celo za 24 točk; večina preostale razlike izhaja iz odpravljivih pomanjkljivosti pri gradnji grafa, edina resnična omejitev pa so količine, ki jih Cypher ne more izračunati. To je pomembno, ker lahko podatki o stavbah obstajajo v odprti, skupni obliki grafa in kljub temu odgovarjajo na vprašanja skoraj tako dobro kot neposredno branje lastniške datoteke, kar je koristno za vsakogar, ki gradi naravnojezikovna orodja nad gradbenimi podatki ali deluje na področju Linked Building Data. Kar ostaja, je pretežno inženirsko delo: izboljšanje pretvorbe datotek ter dodajanje zmožnosti, da sistem vrednost izračuna namesto da jo le poišče, tako da lahko odgovori tudi na vprašanja o geometriji in materialih, ki jih graf sam po sebi ne more ponuditi.

• Ključne besede: magistrska dela, gradbeništvo, ..., ..., ...