Analyzing Temporal Trends in Public Health Topics Using Association Rule Mining

Public health requires an understanding of the complex and interdependent nature of a variety of topics across multiple disciplines. The growing volume of the public health literature provides a timely opportunity to gain valuable insights into the historical development of public health. This study aims to identify emerging trends in public health topics and examine their relationships over the years indicated by the pertinent literature.

We applied association rule mining to MEDLINE citations on public health, retrieved from PubMed (1985~2005). From each citation formatted in XML, we extracted MeSH (Medical Subject Headings) indexing terms and used them in generating association rules with the Apriori algorithm. To facilitate computation and evaluation of association rules, we grouped the indexing terms into broader categories using the MeSH hierarchy and UMLS (Unified Medical Language System) semantic types. The results of applying the algorithm allowed us to evaluate which terms were most frequently used in the citations and how those terms were related each other. To investigate how frequencies and relationships among different terms changed over time, we conducted an analysis using time-partitioned citations. We examined association rules from each partition and compare the rules generated from previous partitions. The focus of the analysis was to understand how a variety of public health topics, represented by association rules, changed over different time frames. Assessment of the results showed that our approach provided potentially interesting and useful trends in public health topics and their association patterns that would engage interests of diverse public health professionals.