The current cybersecurity landscape is intricate and severe. Accurately extracting cybersecurity entities from threat intelligence and clarifying their relationships is of great significance for strengthening cybersecurity defense. However, the complexity of relation expressions and the prevalence of relation overlapping in natural language texts pose significant challenges to efficient and accurate information extraction. To tackle these challenges, this paper proposes an enhanced model based on the OneRel model. Specifically, the model incorporates a feature extraction module that combines IDCNN and BiGRU to capture local and global features effectively. It also integrates an Attention Mechanism to enhance the model’s focus on critical relational information. Experimental results demonstrate that the improved model achieves F1 scores of 0.7794 and 0.6885 on the Baidu relation extraction dataset and our constructed cyber threat intelligence entity relation dataset, respectively, representing improvements of 2.22% and 2.95% over the original model. This study shows that the enhanced model significantly improves the learning of contextual information in Chinese text sentences, effectively boosting the accuracy and robustness of relation extraction.