Användning av avancerad Retrieval-Augmented Generationför att förbättra sökningar i tvåspråkigakunskapshanteringssystem
Year: 2025
Type: dissertation
Abstract: This Master of Science thesis investigates how Retrieval-Augmented Generation (RAG) technology can be improved using reasoning techniques like multi-hop and re-ranking. The project was conducted in collaboration with Nexer. The goal of the project was to create a chatbot that answers questions based on internal documents, with support for both English and Swedish. The system was developed to be run on premise and contains components for document retrieval, language detection, and multi-hop reasoning. Two variants of RAG implementations was compared: one simple and one with reasoning. Performance was measured in terms of answer quality and retrieval accuracy. The results show that the reasoning based RAG implementation performs better than the simple RAG implementation when asked complex questions that require multiple documents to be answered. The system can be integrated with user friendly tools like Microsoft Teams and have great potential for further development.
Authors: Eriksson, Hugo, Falk, Erik
Institution: Örebro University
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7080968353