My research is aimed at understanding verbal and non-verbal human communication in mass media and its effects by scalable, computational, and data-driven approaches. A big challenge in modern media studies lies in the sheer amount of data in various forms and multiple modalities. To quantitatively characterize communicative activities in this space, I use automated or semi-automated methods that scale indefinitely. In particular, my research emphasizes the visual dimension of human communication, to understand how we communicate via visual means such as facial expressions or gestures using images or videos. I develop the state-of-the-art computer vision and machine learning techniques to automatically recognize contents from large-scale visual data. I further examine what people intend to mean by visuals and how specific visuals are crafted to achieve various communicative goals. The specific problems that I have worked on include social perception, media bias detection, cultural and behavioral diffusion on social media, and emotion analysis of TV news programs.