One of the main approaches to obtaining data related to cement environmental load is using entity relation extraction methods to extract information from texts. However, cement production-related texts contain numerous technical terms, exhibit severe relation overlapping issues, and pose challenges for traditional joint extraction models in capturing textual structural information, leading to low extraction accuracy. To address these challenges, we propose the Graph Attention Extraction with Entity to Relation (GatE2R) method. This method employs a “self-to-self” annotation mechanism, treating entities as a special type of relation to reduce the number of labels. Additionally, it converts text into a graph structure and applies attention aggregation using two adjacency matrices, effectively leveraging textual structural information. Experiments conducted on our self-constructed cement environmental load text dataset demonstrate that the GatE2R method is a reliable framework for cement environment load information extraction.