融合多维特征的医学知识图谱分步实体对齐方法

目的：提出一种融合多维特征的医学知识图谱分步实体对齐方法，从电子病历和网络资源中抽取垂体瘤相关疾病、症状数据，进行实证研究。方法：首先进行尾实体对齐，通过训练Word2Vec和BERT模型获得实体的语义特征，使用三元组训练翻译模型得到实体结构特征，利用Jaccard相似度计算字符特征，利用分类模型进行特征学习和预测；然后进行头实体对齐，利用实体的属性相似性和结构相似性构建头实体对齐模型。结果：尾实体对齐模型的F1值为99.58％，头实体对齐模型的F1值为97.32％，说明所选择的特征可以很好地表示实体，模型具有良好的对齐效果。结论：目前关于医学知识图谱的实体对齐模型研究仍处于起步阶段，融合多维特征的医学知识图谱分步实体对齐方法是对现有医学知识图谱构建方法的重要补充。

A hierarchical entity alignment method for medical knowledge graph based on multi-dimensional features

Objective To propose a hierarchical entity alignment method for medical knowledge graph based on multi-dimensional features, with the data of diseases and symptoms related to pituitary tumor extracted from electronic medical record and network to conduct the empirical research. Method First, tail entity alignment was performed to obtain the semantic features of entities by training Word2Vec and BERT models, to obtain entity structural features using triple training translation model, to calculate character features using Jaccard similarity, and to perform feature learning and prediction using classification model. Then the head entity alignment was performed, which was constructed by using the attribute similarity and structural similarity of the entities. Result The experimental results showed that the F1 value of the tail entity alignment model was 99.58%, and the F1 value of the head entity alignment model was 97.32%, indicating that the selected features could represent the entities well, and the model had a good alignment effect. Conclusion At present, the research of entity alignment model for medical knowledge graph is still in the initial stage and the hierarchical entity alignment method for medical knowledge graph based on multi-dimensional features is an important supplement to the existing medical knowledge graph construction methods.