<p>With the technological transition toward renewable energy, the global demand for lithium resources has surged, highlighting the urgent need for efficient exploration of pegmatite-type lithium deposits. However, most prior research knowledge on pegmatite-type lithium deposits is embedded in vast amounts of literature in natural language form, rendering this unstructured knowledge difficult to utilize directly and significantly prolonging the decision-making process for deposit exploration. To address the knowledge gap in the exploration of pegmatite-type lithium deposits, this paper proposes a knowledge graph-guided intelligent question-answering (KGG-IQA) method. First, the proposed method establishes relevant entities and relationships in the knowledge graph by collecting geological data from research reports and academic papers. Then, the method learns hidden representations of entities and relationships to identify and remove corpus noise to avoid the degradation of answer quality due to erroneous corpora. Finally, the proposed method transforms the user’s input query into a vectorized representation and compares it with stored knowledge vectors using cosine similarity. This process retrieves the top-k most relevant text blocks, generating contextually relevant and accurate answers through a large language model within a retrieval-augmented generation (RAG) framework. The proposed method is validated through qualitative evaluation by expert anonymous scoring and quantitative evaluation using METEOR and BERTScore metrics. Experimental results demonstrate that the proposed method significantly outperforms existing models in terms of accuracy and cost-effectiveness, substantially enhancing the accessibility of unstructured knowledge and supporting strategic decision-making for pegmatite-type lithium deposit exploration.</p>