CS G220 Project Report : Ideology Detection Engine

In this report, I present an early iteration of a system intended to detect the ideological bent of an arbitrary block of text along the scale from liberal to conservative. This is attempted via the use of an Adaboostgenerated ensemble of decision stumps which split upon the presence or absence of a given n-gram in the text. The training data consists of floor statements taken from the U.S. Congressional Record, paired with ideological scores for each elected representative, such that each statement is labeled according to its speaker’s ideology.