The representation of knowledge based on first-order logic captures the richness of natural language and supports multiple probabilistic inference models. Although symbolic representation enables quantitative reasoning with statistical probability, it is difficult to utilize with machine learning models as they perform numerical operations. In contrast, knowledge embedding (i.e., high-dimensional and continuous vectors) is a feasible approach to complex reasoning that can not only retain the semantic information of knowledge, but also establish the quantifiable relationship among embeddings. In this paper, we propose a recursive neural knowledge network (RNKN), which combines medical knowledge based on first-order logic with a recursive neural network for multi-disease diagnosis. After the RNKN is efficiently trained using manually annotated Chinese Electronic Medical Records (CEMRs), diagnosis-oriented knowledge embeddings and weight matrixes are learned. The experimental results confirm that the diagnostic accuracy of the RNKN is superior to those of four machine learning models, four classical neural networks and Markov logic network. The results also demonstrate that the more explicit the evidence extracted from CEMRs, the better the performance. The RNKN gradually reveals the interpretation of knowledge embeddings as the number of training epochs increases.

基于一阶逻辑的知识表示捕获了自然语言的丰富性，并支持多种概率推理模型。尽管符号表示可以统计概率进行定量推理，但是由于机器学习模型执行数字运算，因此很难利用它们。相反，知识嵌入（即，高维连续向量）是一种可行的复杂推理方法，不仅可以保留知识的语义信息，而且可以建立嵌入之间的可量化关系。在本文中，我们提出了一种递归神经知识网络（RNKN），它将基于一阶逻辑的医学知识与递归神经网络相结合，以进行多疾病诊断。在使用人工注释的中文电子病历（CEMR）对RNKN进行有效培训之后，就可以学习面向诊断的知识嵌入和权重矩阵。实验结果证明，RNKN的诊断准确性优于四种机器学习模型，四种经典神经网络和马尔可夫逻辑网络。结果还表明，从CEMR中提取的证据越明确，性能越好。随着训练时期的增加，RNKN逐渐揭示了知识嵌入的解释。