Latent Dirichlet Allocation Topic Models Applied to the Centers for Disease Control and Prevention's Grant Portfolio

• Activity Number: 68 - Government Health Statistics
• Type: Contributed
• Date/Time: Sunday, July 30, 2017 : 4:00 PM to 5:50 PM
• Sponsor: Government Statistics Section
• Author(s): Matthew Eblen* and Robin Wagner
• Companies: Centers for Disease Control and Prevention and Centers for Disease Control and Prevention
• Keywords: Public Health ; Natural Language Processing ; Machine Learning ; Topic Models ; Latent Dirichlet Allocation

Abstract

In fiscal year 2015, the Centers for Disease Control and Prevention (CDC) administered over $5 billion in grants to institutions across the United States and the world. Each grant was administered by one of CDC's 13 Centers, Institutes or Offices (CIOs), each of which has responsibility for different areas of public health. The scope and content of those grants varied widely. This paper explores the use of natural language processing and machine learning to uncover common themes, or topics, in the content of those grants. Specifically, a Latent Dirchlet Allocation (LDA) topic model was applied to a corpus of CDC grant abstracts, resulting in topical word clusters that categorized CDC's recent investments in public health. The LDA topic estimates for grants aligned well with the subject areas of their respective CIOs, but also demonstrated that there were significant areas of overlap and mutual interest, as multiple CIOs administered grants with similar topical content. Trends in funding for the different topics were also examined, resulting in identification of public health areas that have been receiving increased attention and investment over time.