During coal mine production, dispatch room phone recordings contain a wealth of critical information, but existing analytical methods are limited in fully extracting their value. To address this, this paper proposes an automated analysis method based on GraphRAG technology, aiming to conduct event correlation analysis on the speech recognition transcripts from the coal mine dispatch room. This approach enables more efficient identification and resolution of potential safety hazards, thereby improving overall mine safety. Experimental results demonstrate that this method significantly enhances the accuracy and timeliness of hazard identification, showing broad application prospects.