Enhancing information accessibility in infrastructure construction: Using retrieval-augmented generation for efficient parsing of information requirements
Year: 2025
Type: dissertation
Abstract: Information management plays a central role in modern infrastructure projects, requiring the efficient interpretation of various project documents such as requests for proposals and requirement specifications. Identifying and summarizing the information requirements within these documents can be time-consuming and prone to error. Retrieval-Augmented Generation (RAG) combines the generative capabilities of Large Language Models (LLMs) with vector-based retrieval, potentially improving information access and interpretation in the context of infrastructure projects. This thesis investigates whether a simple hybrid RAG implementation can serve as a useful tool for the automatic retrieval and summarization of information requirements. A RAG system is implemented in which project documents are converted into plain text, divided into smaller chunks, vectorized using an embedding model, and stored in a database. Based on this data, the language model is used to answer user queries. The study is guided by three research questions: (1) How effectively can a hybrid RAG system retrieve relevant information requirements from infrastructure project documents? (2) Can the system accurately summarize and interpret the retrieved content? (3) What are the main challenges and limitations of such a system? The system is tested on real-world datasets from infrastructure projects, and its performance is evaluated accordingly. The results suggest that the RAG approach shows potential in supporting information modeling tasks, particularly in preprocessing and extracting relevant content from large documents. However, challenges remain, especially in retrieval accuracy and the language model’s ability to consistently utilize the retrieved information. The findings offer a foundation for future research and the further development of RAG-based methods for information management in infrastructure projects.
Author: Junttu, Jaakko
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7132874949