Search
Program Calendar
Browse By Day
Browse By Time
Browse By Person
Browse By Room
Browse By Division
Browse By Session Type
Search Tips
Personal Schedule
Sign In
Traditional forms of film criticism have relied on subjective interpretation of, at most, only a few films at a time. These limitations are a product of the very real limitations on human attention that prevent scholars from watching every movie that has been released. As such, we propose a computational approach for estimating the latent political positions, what we refer to as \textit{film slant}, of over 5 million films. To do so, we adapt a Bayesian item response theory model for large sparse matrices intended for roll call analysis and apply it using an Expectation-maximization algorithm to trace user data in the form of film reviews (N=~20,000,000). We find that this method is valid, both in regards to the ideological positions of films, but also their viewers. We finally turn our attention to some substantive results and demonstrate, through an analysis of the dynamics of the central tendency of our set of films, that the film industry does not have any specific liberal bias, but rather partakes in a form of negative polarization in which the system responds in opposition to the ideological position of the sitting American president.
Sean Fischer, The Annenberg School for Communication, U of Pennsylvania
Yphtach Lelkes, University of Pennsylvania