Integrating Knowledge Graph with Large Language Models for Multi-hop Question Generation

Abstract

Question generation (QG) is a fundamental task in natural language processing that involves generating fluent and grammatically correct questions from a given input context, optionally conditioned on an answer. Multi-hop question generation (MHQG), a more complex variant, requires reasoning over multiple pieces of information across diverse contexts to formulate coherent questions. In this work, we propose Knowledge Graph for Question Generation (KG4QG), a novel framework that integrates knowledge graphs with large language models to address the challenges of MHQG. Our approach constructs knowledge graphs from input contexts, encodes them using Graph Attention Networks (GAT), and leverages Sentence Transformers for contextual text embeddings. These enriched representations are then fed into large language models—specifically BART and T5—for multi-hop question generation. We evaluate KG4QG on the HotpotQA dataset, demonstrating that our method achieves superior performance compared to existing state-of-the-art approaches, highlighting the effectiveness of combining structured knowledge and pre-trained language models for complex question generation tasks.

Resources

Details