基于二分网络表示学习的医学实体关系预测研究

Purpose/Significance To combine the research of network representation learning and link prediction for exploring their application in mining potential medical entity relationships, thereby offering a novel perspective for medical knowledge discovery. Methods/Process The literature abstracts were obtained from the PubMed database, and disease and drug treatment information within these abstracts was identified using the Subject-Action-Object (SAO) semantic mining scheme. Drug entities and disease entities were extracted, and a drug-disease bipartite network was constructed. The network structure and node characteristics were analyzed comprehensively using methods of social network analysis, network representation learning, and machine learning to uncover potential connections between medical entities. Results/Conclusion Random forest has the best effect, effectively revealing the association knowledge between drugs and diseases, and demonstrating the practical significance of the research method through the verification of the prediction results.