LLM based automatic relation between cost domain descriptions and IFC objects

Effective cost estimation for tendering plays a critical role in the building construction process, enabling efficient investment management and ensuring successful execution of the construction phase. The current practice involves the classification of building items, extracting all the quantities of the latter, collecting pricing information from construction price list documents and manually relate these data to the building items. The objective of this paper is to support cost estimation activity by developing a tool that automates the process of assigning a cost domain description to IFC-based BIM building objects, in such a way as to minimize the human error when manually performing this activity and speed up the process. To handle the textual dataset, the authors propose a prompt-based framework, testing Mistral-7b language model to querying cost domain descriptions with data in IFC format. This approach is applied to two domains, each characterized by different semantics.