Statistical Topic Models for Science and Innovation Policy

• Activity Number: 5
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Hanna Wallach*+
• Companies: University of Massachusetts at Amherst
• Keywords: statistical topic models ; priors ; Bayesian inference ; science and innovation policy ; computational tools ; computational social science

Abstract

Science and innovation policy is concerned with identifying and selecting high-impact educational, financial, and political actions. In order to make truly data-driven decisions, policy-makers need quantitative tools for analyzing massive, complex collections of information. I will discuss the development of such tools. I will concentrate on a class of models known as statistical topic models, which automatically infer groups of semantically-related words (topics) from word co-occurrence patterns in documents. These topics can be used to detect emergent areas of innovation, identify communities, and track trends across languages. Until recently, most statistical topic models relied on two unchallenged prior beliefs. I will explain how challenging these beliefs increases robustness to the skewed word frequency distributions common in text. I will also talk about a) highly-specialized, fine-grained topic models that yield high-quality topics, as evaluated by human domain experts, and b) topic models that identify latent issue-based voting coalitions within the US senate via bill content and roll call senator voting records.