سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / سومین کنفرانس بین المللی مهندسی و علوم کامپیوتر و اولین کنفرانس ملی سامانه های هوشمند، کلان داده و رایانش پیشرفته

GraphRAG: Next-Generation Graph-Augmented Retrieval-Generation Framework for Secure IoT Intelligence and Large-Scale Data Analytics

نویسندگان :

Mahdi Bagheri¹ Behrang Barekatain² Fezeh Salimi tarary³

1- دانشگاه آزاد اسلامی واحد نجف آباد 2- دانشگاه آزاد اسلامی واحد نجف آباد 3- دانشگاه آزاد اسلامی واحد نجف آباد

کلمات کلیدی :

Keywords—GraphRAG،Knowledge Graphs،Retrieval-Augmented Generation

چکیده :

Abstract— Large Language Models (LLMs), despite remarkable advances in natural language processing, still face critical limitations in enterprise, big data, and Internet of Things (IoT) environments, including hallucinated outputs, static post-training knowledge, limited context windows, and vulnerability to prompt injection attacks. Existing Retrieval-Augmented Generation RAG methods rely on dense vector similarity and lack support for structured semantic reasoning in multi-hop inference. This research introduces GraphRAG, a novel framework that integrates Knowledge Graphs (KGs) with multimodal LLMs to enhance retrieval accuracy, semantic reasoning, and system security. GraphRAG implements a three-layer architecture comprising entity and relation extraction from textual data, semantic knowledge graph construction, and hybrid retrieval that combines vector-based similarity with graph-based context expansion. Security modules based on graph validation and homomorphic encryption are incorporated to mitigate prompt injection and protect sensitive data. Experimental evaluations on WikiHop and a SmartIoT dataset demonstrate that GraphRAG improves retrieval accuracy from 66.2% to 92.8% and contextual faithfulness from 0.74 to 0.91, while incurring only a 17.7% increase in computational latency. Ablation analysis confirms that optimal coefficient balancing delivers highest combined performance across accuracy, faithfulness, and latency metrics.