KG-RAG: Bridging the Gap Between Knowledge and Creativity This paper introduces a KG-RAG (Knowledge Graph-Retrieval Augmented Generation) pipeline to enhance the knowledge capabilities of Large Language Models (LLMs) by integrating structured Knowledge Graphs (KGs) with LLMs. The KG-RAG pipeline constructs a KG from unstructured text and performs information retrieval over the graph to perform KGQA (Knowledge Graph Question Answering). The retrieval methodology leverages a novel algorithm called Chain of Explorations (CoE), which benefits from LLMs reasoning to explore nodes and relationships within the KG sequentially. Preliminary experiments on the ComplexWebQuestions dataset demonstrate notable improvements in the reduction of hallucinated content and suggest a promising path toward developing intelligent systems adept at handling knowledge-intensive tasks. The KG-RAG pipeline consists of three stages: Storage, Retrieval, and Answer Generation. In the Storage stage, unstructured text is converted into a structured KG by extracting triples. In the Retrieval stage, the KG is explored using CoE to find relevant information. In the Answer Generation stage, the LLM uses the retrieved information to generate accurate and contextually relevant answers. The KG-RAG pipeline addresses the challenges of hallucination, catastrophic forgetting, and granularity in dense retrieval systems. It improves upon existing RAG methods by utilizing dynamically updated KGs, addressing information hallucination through more granular and context-sensitive retrieval processes. The KG-RAG pipeline was evaluated on the ComplexWebQuestions (CWQ) dataset, showing promising results in terms of accuracy and reduced hallucination rates compared to other methods. The results indicate that KG-RAG is more adept at adhering to factual content and reducing the generation of unsupported content. However, the financial constraints of the experiments limited the use of all web snippets and the testing on the entire development split of the CWQ dataset. Future research could expand upon this work as budget allowances increase or as the computing costs of LLMs decrease. The integration of structured knowledge into the operational framework of LMAs through KGs represents a significant paradigm shift in how these agents store and manage their information. This integration can significantly reduce the occurrence of hallucinations in LMAs, as it ensures that these agents rely on explicit information rather than generating responses based on knowledge stored "implicitly" in their weights. Additionally, KGs enable LMAs to access vast volumes of accurate and updated information without the need for resource-intensive fine-tuning.KG-RAG: Bridging the Gap Between Knowledge and Creativity This paper introduces a KG-RAG (Knowledge Graph-Retrieval Augmented Generation) pipeline to enhance the knowledge capabilities of Large Language Models (LLMs) by integrating structured Knowledge Graphs (KGs) with LLMs. The KG-RAG pipeline constructs a KG from unstructured text and performs information retrieval over the graph to perform KGQA (Knowledge Graph Question Answering). The retrieval methodology leverages a novel algorithm called Chain of Explorations (CoE), which benefits from LLMs reasoning to explore nodes and relationships within the KG sequentially. Preliminary experiments on the ComplexWebQuestions dataset demonstrate notable improvements in the reduction of hallucinated content and suggest a promising path toward developing intelligent systems adept at handling knowledge-intensive tasks. The KG-RAG pipeline consists of three stages: Storage, Retrieval, and Answer Generation. In the Storage stage, unstructured text is converted into a structured KG by extracting triples. In the Retrieval stage, the KG is explored using CoE to find relevant information. In the Answer Generation stage, the LLM uses the retrieved information to generate accurate and contextually relevant answers. The KG-RAG pipeline addresses the challenges of hallucination, catastrophic forgetting, and granularity in dense retrieval systems. It improves upon existing RAG methods by utilizing dynamically updated KGs, addressing information hallucination through more granular and context-sensitive retrieval processes. The KG-RAG pipeline was evaluated on the ComplexWebQuestions (CWQ) dataset, showing promising results in terms of accuracy and reduced hallucination rates compared to other methods. The results indicate that KG-RAG is more adept at adhering to factual content and reducing the generation of unsupported content. However, the financial constraints of the experiments limited the use of all web snippets and the testing on the entire development split of the CWQ dataset. Future research could expand upon this work as budget allowances increase or as the computing costs of LLMs decrease. The integration of structured knowledge into the operational framework of LMAs through KGs represents a significant paradigm shift in how these agents store and manage their information. This integration can significantly reduce the occurrence of hallucinations in LMAs, as it ensures that these agents rely on explicit information rather than generating responses based on knowledge stored "implicitly" in their weights. Additionally, KGs enable LMAs to access vast volumes of accurate and updated information without the need for resource-intensive fine-tuning.