Evaluating the relevance of health-related topics using three similarity measures

Abstract

This study evaluated the effectiveness of three similarity measures—Cosine similarity, Pearson correlation, and Euclidean distance—in assessing health-related topics on MedlinePlus. The focus was on four health topic subcategories: mental health, children, teenagers, and older adults. Using adjacency matrices of graph theory and the three similarity measures, the study found that both Cosine and Pearson correlation measures were more empirically robust than the Euclidean distance measure. Notably, the alignment in findings from Cosine and Pearson correlation suggests their potential combined use in future research as complementary strategies. To validate the findings, hypothesis testing showed that Cosine and Pearson correlation were significantly effective in identifying similar health topics and distinguishing between different semantic subgroups, whereas Euclidean distance showed limitations. These insights guide the application of adjacency matrices and the selection of suitable similarity measures to evaluate semantic linkages in health topics, enhancing relevance recognition and supporting classification in medical domains.