Enhancing QA on Technical Document with Hybrid, and Graph-Based RAG Approaches

This study evaluates four Retrieval-Augmented Generation (RAG) architectures-Hybrid, Graph-Based, and Graph+Hybrid Rerank-on domain-specific technical documentation. We investigate whether knowledge graphs enhance accuracy and reasoning in complex queries. Experimental results show that while Dense retrieval provides a fast baseline (۶۰% accuracy), Graph-based methods significantly improve contextual reasoning (۸۵%), and Graph+Hybrid Rerank achieves the highest accuracy (۹۵%) at the cost of higher latency. Findings highlight the trade-off between speed and precision, offering a principled framework for selecting RAG strategies in knowledge-intensive applications.

کلیدواژه ها:

Retrieval-Augmented Generation (RAG) ، Knowledge Graph ، Vector-based Retrieval ، Technical Documentation ، Question Answering ، Semantic Search ، Hybrid Architecture ، Information Retrieval ، Context-aware Reasoning