基于多路链接预测推理的中医脏腑辨证研究

Research on Viscera Syndrome Differentiation in TCM Based on Multiple Link Prediction Reasoning

目的 针对知识图谱链接预测中缺少多关系预测推理的问题,提出基于规则+马尔可夫逻辑网(MLN)的多路链接预测推理算法,为中医临床提供辅助决策支持。方法 以中医脏腑辨证为研究背景,构建基于Neo4j图数据库的知识图谱,包括1 263个实体节点和4 105个语义关系。基于改进规则的知识推理设置初始权重,并使用MLN和吉布斯采样(Gibbs sampling)训练权重,以完成输入任意个数四诊数据得到证候推理结果的链接预测任务。结果 对中医脏腑辨证的70个证候进行推理,结果AUC值为98.6%,精确度为98.6%,排序分为0.297,较基于传统规则链接预测算法的精确度高4.3%,具有更准确的推理结果。结论 该模型能较好完成多对多复杂路径关系的链接预测任务,同时实现中医脏腑辨证的四诊合参及个性化推荐功能,辅助中医临床诊疗。

Objective To propose a multi-link prediction reasoning algorithm based on rule + Markov logical network(MLN) in view of the lack of multi-relation predictive reasoning in knowledge graph link prediction;To provide auxiliary decision support for TCM clinic.Methods Taking TCM viscera syndrome differentiation as the research background,a knowledge graph based on Neo4j graph database was constructed,including 1 263 entity nodes and 4 105 semantic relationships.The initial weights were set based on improved rules of knowledge reasoning,and the weights were trained by MLN and Gibbs sampling to complete the link prediction task of syndrome reasoning results obtained by inputting any multiple four-diagnosis data.Results The syndrome reasoning was carried out for 70 syndromes of TCM syndrome differentiation,the AUC was 98.6%,the accuracy was 98.6%,and the ranking score was 0.297,which was 4.3% higher than the accuracy of the traditional rule link prediction algorithm and had more accurate reasoning results.Conclusion The model can better complete the link prediction task of multi-pair and multi-complex path relationship,and realize the combination of four diagnostic methods of TCM viscera syndrome differentiation and the personalized recommendation function to assist traditional Chinese medicine clinical diagnosis and treatment.