现有中医药方推荐方法多通过构建症状-草药、症状-症状和草药-草药的多图结构表达症状与草药间的多维关系，但大多忽略了症状和草药间的语义交互，且缺乏有效的多视图特征融合策略导致重要信息提取不充分以及忽视症状相对重要性导致证候归纳不准确的问题。为此，提出语义知识整合与多视图融合的中医药方推荐模型。首先，引入BERT（bidirectional encoder representations from transformers）模型作为节点词的语义编码器，学习症状和草药间的语义交互信息，增强模型对节点语义特征的表达能力。其次，设计多视图特征融合模块，利用多头注意力机制和低秩矩阵分解技术生成视图间的特征转换矩阵，自适应动态融合不同视图的信息。最后，采用注意力机制准确归纳特定症状集的综合证候，以优化中医药方推荐效果。实验结果表明，本文模型在公开数据集上的关键指标F₁@5、F₁@10、F₁@20较最优基线模型分别提升了10.27%、15.10%和17.29%，证明了其有效性。

The existing traditional Chinese medicine prescription recommendation methods mostly express the multi-dimensional relationship between symptoms and herbs by constructing symptom herb, symptom symptom and herb herb multi graph structures, but mostly ignore the semantic interaction between symptoms and herbs, and the lack of effective multi view feature fusion strategies leads to insufficient extraction of important information and the neglect of the relative importance of symptoms leading to inaccurate syndrome induction. Therefore, a traditional Chinese medicine prescription recommendation model based on semantic knowledge integration and multi view integration was proposed. First, the BERT(bidirectional encoder representations from transformers) model was introduced as the semantic encoder of node words to learn the semantic interaction information between symptoms and herbs, and enhance the model's ability to express node semantic features. Secondly, multi view feature fusion module was designed, which used multi head attention mechanism and low rank matrix decomposition technology to generate feature conversion matrix between views, and adaptively and dynamically fused information of different views. Finally, attention mechanism was used to accurately summarize the syndrome of specific symptom set to optimize the effect of prescription recommendation. The experimental results show that the key indicators of this model on the public dataset F₁@5, F₁@10, F₁@20 is 10.27%, 15.10% and 17.29% higher than the optimal baseline model, which proves its effectiveness.