POLITICAL OPINION IDENTIFICATION, MINING AND RETRIEVAL

Open Access
Author:
Zhu, Lei
Graduate Program:
Information Sciences and Technology
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 28, 2010
Committee Members:
  • Prasenjit Mitra, Thesis Advisor
  • Burt Monroe Iii, Thesis Advisor
Keywords:
  • ideology score
  • political opinion mining
Abstract:
I provide a critical literature review on Computational Political Science in this the- sis, which summarizes studies of political science issues utilizing computational tech- niques. Text analysis and Network analysis, the two main subfields in computational political science are discussed in detail, and the usage of miscellaneous computational techniques in political science is also addressed. I present my studies on the problem of Political Spectrum Analysis, namely text- based ideal point estimate, in Chapter three as an example of computational political science. Political Spectrum refers to a multidimensional opinion space where each geometric axis models one political dimension. Political opinion mining shares some characteristics with product reviews mining [39] [14] while introducing new challenges to opinion identification, modeling and representation. The study starts from the congressional political domain. I show the importance of multidimensional opinion representation in the congressional context combining domain knowledge and results from three different dimensionality analysis methods. Several regression models are trained to get ideology scores from the text, based on both Bag-of-words feature sets and Topic-based feature sets. I also transfer to the civic political domain by studying a tagged blog space with the learned regression models from the congressional domain. Real world applications of both political opinion mining and political opinion retrieval are discussed in the last chapter and several user scenarios are proposed to conclude the contribution of my studies and reflect future potential.