Enhancing Retrieval-Augmented Generation Models with Knowledge Graphs: Innovative Practices Through a Dual-Pathway Approach

This manuscript delves into the augmentation of Retrieval-Augmented Generation (RAG) through Knowledge Graphs (KG), aimed at elevating the performance of Natural Language Processing (NLP) tasks. Despite the remarkable strides made by Large Language Models (LLMs) in the domain of language comprehension and generation, challenges persist in handling tasks requiring granular or specialized domain knowledge. To address this issue, researchers have proposed the RAG system, which enhances task performance and interpretability by amalgamating retrieval modules with LLMs. However, current RAG systems still face deficiencies in managing retrieval noise and hallucination issues. The article proposes a novel dual-pathway approach, integrating structured Knowledge Graph data into the retrieval module of RAG, to refine retrieval quality and yield more accurate and coherent outputs. Experimental validation demonstrates the efficacy of this approach in enhancing the accuracy and reliability of generated content, particularly in controlling hallucinations and bolstering the output's reliability when processing lengthy text inputs. Moreover, this methodology offers significant flexibility and customizability, facilitating adjustments in retrieval patterns and output formats according to the diverse requirements of different tasks, heralding widespread applicability in artificial intelligence-powered question-answering systems and text comprehension tasks.