To enhance the accuracy of extracting complex，multi-level causal relationships in aquatic disease control，this study introduces a bidirectional focus enhancement mechanism （BFM） and proposes the RoBFM model （RoBERTa-wwm-ext-large integrated with bidirectional focus enhancement mechanism）.The model integrates the RoBERTa-wwm-ext-large pre-trained language model to generate high-quality word embeddings and leverages Bi-directional long short-term memory （BiLSTM） networks to strengthen long-distance dependency modeling.The core innovation of RoBFM resides in the BFM mechanism，which employs forward and backward weight allocation to explicitly decouple and focus on causal semantic roles.To systematically evaluate the model’s performance，this paper conducts baseline comparison and ablation experiments on the public dataset DuIE and the self-constructed，domain-specific dataset DLOU-CRE.Experimental results demonstrate that the proposed model surpasses existing baselines in terms of F₁ score and other metrics on both datasets.Notably，the model achieves an F₁ score of 68.23% on the DLOU-CRE dataset，demonstrating that the proposed method is more effective for causal relationship extraction in the prevention and control of aquatic animal diseases.