Effective disease prevention and control are fundamental for ensuring the sustainable and healthy aquaculture industry development. To achieve precise and efficient disease risk management, constructing a high-quality knowledge graph capable of integrating heterogeneous and multisource knowledge is essential. Within such a system, knowledge completion serves as a critical process in refining the knowledge structure and as a core component in enhancing the reliability and robustness of the overall knowledge infrastructure. However, the accuracy of knowledge graph completion remains low, primarily because of the widespread “different strategies for the same disease” phenomenon and the limited efficiency of existing methods in extracting knowledge from long texts. To address these issues, we propose the DeepSeek-based dynamic knowledge graph completion framework for aquaculture disease prevention and control (DyAquaKG). By deeply integrating large language models with domain-specific knowledge, this method substantially increases knowledge graph completion accuracy. First, to mitigate the “different strategies for the same disease” issue, we developed an entity classification and annotation strategy based on the damaged object. By integrating a DeepSeek model optimized with vertical domain prompt templates and a global–local dual-stage collaborative retrieval mechanism, long-text context dependency loss is effectively mitigated, enhancing semantic retrieval accuracy and consistency. To deepen the understanding of the semantic relationships among triples associated with the “different strategies for the same disease” phenomenon, a differentiable logic rule enhancement module is employed to enable advanced knowledge graph completion and reasoning optimization. The experimental results demonstrate that the proposed method substantially outperforms baseline approaches across all evaluation metrics for aquaculture disease prevention and control knowledge graph completion. In the link prediction task, the proposed method achieves a mean rank of 48.6, a mean reciprocal rank of 73.0%, and a Hits@10 score of 87.0%. Overall, the proposed framework offers an effective approach to knowledge graph completion in aquaculture disease prevention and control, and it contributes to advancing the intelligent development of domain-specific knowledge services.
Huang et al. (Sun,) studied this question.