Knowledge Graphs are all you need: Leveraging KGs in Physics Question Answering
Project Overview
The document explores the integration of generative AI in education, particularly focusing on enhancing high school physics learning through a novel methodology that combines knowledge graphs with large language models (LLMs). This innovative approach involves breaking down complex physics questions into logically consistent sub-questions using knowledge graphs, which aims to improve the quality of educational content and significantly enrich the learning experience. The study highlights substantial benefits in terms of clarity and relevance of the answers provided, while also tackling challenges related to model reasoning and the generalization of AI applications across various educational domains. Overall, the findings indicate that generative AI can effectively support students' understanding of complex subjects by making educational content more accessible and tailored to individual learning needs.
Key Applications
Knowledge graph-based question decomposition pipeline
Context: High school physics education, targeted at high school students
Implementation: Utilizing LLMs to generate knowledge graphs from original questions, guiding the generation of logically consistent sub-questions.
Outcomes: Improved fidelity of sub-questions to original questions' logic, enhanced learning experience, and better quality of answers.
Challenges: Limited internal knowledge of LLMs leading to occasional errors in sub-query resolution.
Implementation Barriers
Technical
LLMs may struggle with complex multi-step questions that require logical reasoning and domain-specific knowledge.
Proposed Solutions: Integrate knowledge graphs into the question-answering process to provide a more organized knowledge base.
Data Availability
The need for large amounts of quality training data to fine-tune LLMs.
Proposed Solutions: Develop unique datasets, such as the one created in this study, to enhance LLM training for educational contexts.
Project Team
Krishnasai Addala
Researcher
Kabir Dev Paul Baghel
Researcher
Dhruv Jain
Researcher
Navya Gupta
Researcher
Rishitej Reddy Vyalla
Researcher
Chhavi Kirtani
Researcher
Avinash Anand
Researcher
Rajiv Ratn Shah
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Krishnasai Addala, Kabir Dev Paul Baghel, Dhruv Jain, Navya Gupta, Rishitej Reddy Vyalla, Chhavi Kirtani, Avinash Anand, Rajiv Ratn Shah
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai