Enhancing Retrieval - Augmented Generation with Graph-Based Retrieval and Generative Modeling




Abstract:
This paper presents the design and implementation of a robust RetrievalAugmented Generation (RAG) system that integrates advanced retrieval, ranking, and generative techniques to address knowledge-intensive tasks. The system combines dense retrieval using ChromaDB, metadata-driven keyword extraction with YAKE and KMedoids algorithm for clustering keywords, graph-based retrieval leveraging PageRank, and cross-encoder re-ranking to deliver precise and contextually relevant results. These retrieval outputs are synthesized into high-quality conversational responses using Hugging Face models and Google API. A modular pipeline ensures scalability, seamlessly integrating various retrieval and generative components. Evaluation results demonstrate high retrieval precision, improved recall through graph-based methods, and enhanced response quality through structured prompt engineering. This work highlights the effectiveness of combining diverse techniques in RAG systems, offering a foundation for scalable, reliable, and context-aware applications in domains such as customer support, education, and research.

CITATION:

IEEE format

D. Vujić, A. Njeguš, N. Bačanin Džakula, “Enhancing Retrieval - Augmented Generation with Graph-Based Retrieval and Generative Modeling,” in Sinteza 2025 - International Scientific Conference on Information Technology, Computer Science, and Data Science, Belgrade, Singidunum University, Serbia, 2025, pp. 3-9. doi:10.15308/Sinteza-2025-3-9

APA format

Vujić, D., Njeguš, A., Bačanin Džakula, N. (2025). Enhancing Retrieval - Augmented Generation with Graph-Based Retrieval and Generative Modeling. Paper presented at Sinteza 2025 - International Scientific Conference on Information Technology, Computer Science, and Data Science. doi:10.15308/Sinteza-2025-3-9

BibTeX format
Download

RefWorks Tagged format
Download