Large Language Models (LLM’s) to support ontology learning

At 𝗕𝗜𝗠-𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗲𝗱, we’re always exploring how cutting-edge technology can help structure, connect, and make sense of information. One of our latest internal experiments? Using 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗟𝗟𝗠’𝘀) to support 𝗼𝗻𝘁𝗼𝗹𝗼𝗴𝘆 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴.

🔧 Over the past few months we:
– Selected a promising academic approach from recent research,
– Set up pipelines to run experiments using  𝗿𝗲𝗮𝗹 𝗶𝗻𝗱𝘂𝘀𝘁𝗿𝘆 𝘀𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝘀 like 𝗡𝗘𝗡 𝟮𝟲𝟲𝟬 and 𝗜𝗠𝗕𝗢𝗥,
– Tested different models and prompting methods (OpenAI, Mistral, DeepSeek, Spacy),
– Created a prototype app to extract ontologies from natural language descriptions,
– And began integration with our own platform, Wistor.

We treated this as an 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝗮𝗹 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 𝘁𝗿𝗮𝗰𝗸—bridging research with practical BIM use cases. The result is a 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗽𝗿𝗼𝗼𝗳-𝗼𝗳-𝗰𝗼𝗻𝗰𝗲𝗽𝘁 that shows how we might bring semantic technologies closer to everyday workflows.

This kind of initiative reflects who we are as a company: always curious, always experimenting, and always aiming to turn complex ideas into real-world solutions.

💡What’s next? We’re now thinking about how to take this further: better criteria, more use cases, and broader accessibility. We’ll keep working on this in the coming period.