Utilizing a RAG-powered LLM for information retrieval in a research repository

Abstract

Though widely use across many research repositories, keyword search may not be sufficient for people who are becoming more familiar with the use of chatbots like ChatGPT. The proposed system will serve as a search engine for the UPM IRS which is a repository for the university’s theses. The system will utilize the vector space model in retrieving documents by directly embedding the user’s query into a vector to be compared to the vectors stored in a vector store by cosine similarity. Retrieval Augmented Generation (RAG) will then be used as the top documents will be given to a large language model (LLM) to create an overview of the top documents. The combination of a semantic retrieval method and a LLM was able to yield a good user experience and relevant results to the users.