Question Answering Using Large Language Models Based on Custom Database
Click the Poster to View Full Screen, Right click to save image
Grant: Kean University and American Council on Education
Mostafa Moamen
CoPIs:
Natalie Olenkovsky, Cymantha Blackmon, Habiba Morsy, Essence Toone
College:
The Dorothy and George Hennings College of Science, Mathematics, and Technology
Major:
Computer Science
Faculty Research Advisor(s):
Ensela Mema, George Avirappattu
Abstract:
This project utilizes Large Language Models (LLMs) to answer questions on specialized topics, focusing on the Carnegie Classification of Institutions of Higher Education (CCIHE). Employing the most prominent LLMs used in the industry, such as GPT by OpenAI and Llama II by Facebook, we develop an application capable of comprehending user queries, retrieving relevant information from existing knowledge and formulating responses.
Beginning with a database containing approximately 300 frequently asked questions and answers, we employ the Retrieval Augmented Generation (RAG) technique to generate responses. This process involves embedding common questions into a vector database, identifying and retrieving the most relevant Q&A pairs to the user query, and employing three large LLMs – GPT 3.5 Turbo, GPT 4, and Llama II – to generate responses based on the retrieved information.
Once developed, the application will be deployed on a website interface constructed using React, a widely used technology for web development. The website will enable administrators to integrate new questions and human-validated responses into the Q&A database, thus enriching the knowledge base and elevating answer quality over time.