Nikhil Shyamlal Yadav, Anand Raj Yadav, Shruti Ravindra Gaikwad, Dr. Vijay Dhawale

DOI: https://doi.org/10.5281/zenodo.24130056

Retrieval-Augmented Generation (RAG) supplements a large language model with evidence retrieved from an external knowledge source. Vector retrieval is effective for semantic similarity, whereas graph retrieval represents entities and explicit relationships that can support multi-hop and corpus-level questions. This paper develops a comparative framework for evaluating vector-only, graph-only, and hybrid vector-graph retrieval under a shared generation setup. The review covers dense passage retrieval, GraphRAG, HippoRAG, LightRAG, and evaluation frameworks for retrieval-augmented systems. A reproducible evaluation protocol is proposed around 50 stratified questions and metrics for answer accuracy, evidence recall, faithfulness, citation coverage, latency, and resource overhead. Because no query logs, model outputs, corpus, or annotation records were supplied for this manuscript, Section 5 uses clearly identified synthetic values solely to demonstrate how a 50-question analysis and its figures would be reported. These illustrative values suggest a plausible trade-off: hybrid retrieval may improve coverage of relational questions while increasing latency and infrastructure complexity. They are not empirical findings and cannot establish superiority. The contribution is therefore a transparent comparative design, a reusable evaluation matrix, and a set of hypotheses for a controlled experiment. A publishable empirical claim requires rerunning the protocol on a documented dataset and replacing all synthetic values with measured results.

Graph database; Hybrid retrieval; Knowledge graph; Retrieval-Augmented Generation; Vector database; Evaluation.

This paper presents a comparative framework for vector-only, graph-only, and hybrid retrieval in Retrieval Augmented Generation. Existing research indicates that vector similarity and graph structure provide comple mentary mechanisms: vector retrieval finds semantically relevant passages, while graph retrieval exposes explicit relations and can support multi-hop or corpus-level ques tions. A controlled evaluation should hold the corpus, generator, prompt, and evidence budget constant while measuring answer accuracy, evidence recall, faithfulness, citation coverage, latency, and resource overhead. The example analysis demonstrates a reporting workflow and does not establish empirical superiority for any architec ture. The appropriate design depends on the target workload. Vector-only retrieval may be sufficient for independent facts and semantic lookup; graph retrieval may be valu able when relations are accurately represented; and hy brid retrieval may justify its additional complexity when questions require both semantic discovery and relational expansion. These expectations must be tested with docu mented configurations, question-level logs, blinded anno tations, and uncertainty estimates. Sharing the query set, permitted source materials, retrieval traces, annotation rubric, and analysis scripts would strengthen replication. The contribution of this study is therefore a reproducible research design and testable hypotheses whose conclu sions remain conditional until the experiment is executed and independently checked.

Asai, A., Wu, Z., Wang, Y., Sil, A., & Hajishirzi, H. (2024). Self-RAG: Learning to retrieve, generate, and critique through self-reflection [Preprint]. arXiv. https://arxiv.org/abs/2310.11511 Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., Metropolitansky, D., Ness, R. O., & Larson, J. (2024). From local to global: A GraphRAG approach to query-focused summarization [Preprint]. arXiv. https://arxiv.org/abs/2404.16130 Es, S., James, J., Espinosa-Anke, L., & Schockaert, S. (2024). RAGAs: Automated evaluation of retrieval augmented generation. In Proceedings of the EACL: System Demonstrations (pp. 150–158). https://doi.org/10.18653/v1/2024.eacl-demo.16 Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., Dai, Y., Sun, J., Meng, W., & Wang, H. (2023). Retrieval-augmented generation for large language models: A survey [Preprint]. arXiv. https://arxiv.org/abs/2312.10997 Guo, Z., Xia, L., Yu, Y., Ao, T., & Huang, C. (2024). LightRAG: Simple and fast retrieval-augmented generation [Preprint]. arXiv. https://arxiv.org/abs/2410.05779 Gutiérrez, B. J., Shu, Y., Gu, Y., Yasunaga, M., & Su, Y. (2024). HippoRAG: Neurobiologically inspired long-term memory for large language models [Preprint]. arXiv. https://arxiv.org/abs/2405.14831 Karpukhin, V., Oguz, B., Min, S., Lewis, P., Wu, L., Edunov, S., Chen, D., & Yih, W.-T. (2020). Dense passage retrieval for open-domain question answering. In Proceedings of EMNLP (pp. 6769–6781). https://doi.org/10.18653/v1/2020.emnlp-main.550 Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-T., Rocktäschel, T., Riedel, S., & Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in neural information processing systems (Vol. 33, pp. 9459–9474). https://arxiv.org/abs/2005.11401 Linders, J., & Tomczak, J. M. (2025). Knowledge graph-extended retrieval augmented generation for question answering. Applied Intelligence, 55, Article 1102. https://doi.org/10.1007/s10489-025-06885-5 Saad-Falcon, J., Khattab, O., Potts, C., & Zaharia, M. (2024). ARES: An automated evaluation framework for retrieval-augmented generation systems. In Proceedings of NAACL-HLT (pp. 338–354). https://doi.org/10.18653/v1/2024.naacl-long.20 Yan, S.-Q., Gu, J.-C., Zhu, Y., & Ling, Z.-H. (2024). Corrective retrieval augmented generation [Preprint]. arXiv. https://arxiv.org/abs/2401.15884

Nikhil Shyamlal Yadav, Anand Raj Yadav, Shruti Ravindra Gaikwad, Dr. Vijay Dhawale (2026). Hybrid Retrieval for Retrieval-Augmented Generation using Vector and Graph Databases. International Journal of Computer Techniques, 13(6), 676–684. ISSN: 2394-2231. DOI: https://doi.org/10.5281/zenodo.24130056