Zero-shot learning for knowledge graph construction and RAG for question answering

In the context of information explosion, extracting valid information from multi-granular redundant data and realizing intelligent processing of unstructured information are core requirements of information extraction. The intelligent question-answering module must balance efficiency and accuracy. This study uses the Wikipedia entries of XX Province power plants as samples to explore knowledge graph construction and the design of a RAG question-answering framework. By integrating chain-prompting design for entity and relation triples extraction, applying post-processing techniques for granularity alignment, and using a triple-augmented hybrid retrieval mechanism to assist answer generation, the method is evaluated. Experimental results show that this approach achieves state-of-the-art (SOTA) accuracy in information extraction across multiple large language models, and the question-answering module significantly outperforms raw text input in terms of answer completeness. The findings provide a reliable knowledge foundation and analytical tools for regional power plant research.