<?xml version="1.0"?>
<metadata xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:dc="http://purl.org/dc/elements/1.1/"><dc:title>Verifying Quantity Takeoff Documents Against BIM Models: A Correspondence Verification Framework with Deterministic, LLMassisted and Knowledge-based Resolution Mechanisms</dc:title><dc:creator>Daulenskytė,	Vilmantė	(Avtor)
	</dc:creator><dc:creator>Turk,	Žiga	(Mentor)
	</dc:creator><dc:subject>master thesis</dc:subject><dc:subject>civil engineering</dc:subject><dc:subject>model to document correspondence verification</dc:subject><dc:subject>cross-lingual semantic matching</dc:subject><dc:subject>semantic table interpretation</dc:subject><dc:subject>building information modeling</dc:subject><dc:subject>material takeoff</dc:subject><dc:subject>LLM-based information extraction</dc:subject><dc:description>Verifying the consistency of quantities between a takeoff document and a BIM model is a contractual obligation. When performed manually, the process is time consuming, prone to human error and leaves no auditable record. Prior research has primarily addressed checking models against formal rules or deriving quantities from models. It has paid less attention to verifying an independently delivered document against an independently delivered model. Current solutions are largely preventive: they establish correspondence during authoring through standards, integrated toolchains and shared vocabularies. This thesis addresses cases in which these mechanisms are absent, as is common in Lithuanian practice. It examines the deliverables as received: a non-machine-readable takeoff and an IFC compliant model in which product identifying fields may be empty or misleading. The thesis investigates how reliably takeoff rows can be matched to model elements under conditions found in actual deliverables, which properties of the deliverables determine this reliability, and how three mechanism families perform on the same task under identical conditions. Following Design Science Research, it develops the Correspondence Verification Framework, which integrates entity matching, language model based extraction and knowledge-based inference within a common pipeline containing one interchangeable resolution component. The framework is demonstrated on three Lithuanian structural projects and evaluated against reference data verified row by row by the author. The evaluation reports precision among resolved rows, the resolution rate over a fixed set of rows, resolution depth and row-level consistency verdicts presented through annotated takeoffs and BCF issues. The results show that correspondence can be established from the delivered content alone, but only at the granularity supported by the deliverables: generally at the level of element types and at the level of individual elements only where identifying marks are provided. The three mechanism families differ more substantially in cost and auditability than in attainable matching quality. The framework provides cost engineers with an automated first-pass check of signed deliverables and provides clients and BIM managers with empirical evidence about which information requirements, particularly a shared identifying mark, make independently delivered models and documents verifiable.</dc:description><dc:publisher>[V. Daulenskytė]</dc:publisher><dc:date>2026</dc:date><dc:date>2026-09-26 08:45:22</dc:date><dc:type>Magistrsko delo/naloga</dc:type><dc:identifier>188716</dc:identifier><dc:identifier>UDK: 004.42:69(043.2)</dc:identifier><dc:identifier>VisID: 183601</dc:identifier><dc:identifier>COBISS_ID: 292909827</dc:identifier><dc:language>sl</dc:language></metadata>
