Agenda setting theory explains how media affects its audi- ence. While traditional media studies have done extensive research on agenda setting, there are important limitations in those studies, including using a small set of issues, running costly surveys of public interest, and manually categorizing the articles into positive and negative frames. In this paper, we propose to tackle these limitations with a computational approach and a large dataset of online news. Overall, we demonstrate how to carry out a large-scale computational research of agenda setting with online news data using ma- chine learning.