Methods Workshop – Groeling/Li/Zhu
DateFebruary 8, 2016
Time3:30pm to 5:00pm
Location
4357 Bunche Hall
Contact
Media news plays a vital role in informing citizens, affecting public opinion, and influencing policy making. The flow of information in society is an important topic in social and political science research, but the sheer volume of news data has made manual analysis exceptionally difficult. In this talk, we will present a joint project on joint image-text parsing, including news story segmentation, topic detection, clustering, and tracking with a hierarchical statistical model called And-Or Graph. Detected topics can show a comprehensive interpretation of the related news events, and topic trajectories can reveal how topics evolve over time. Together, these tools provide an automatic news parsing solution to feed further analysis. This information is further combined with sentiment analysis in tweets.Then we will show two case studies: (i) tracking the 2016 U.S. presidential election; and (ii) analyzing gun violence events. We have built a website (viz2016.com) to present accessible visual summaries of our large-scale election tracking results. We will introduce the website in this talk and show how to use the data. For the gun violence case study, we will discuss our topic tracking and sentiment analysis results. We will also talk about our automated analyses of news visuals, including communicative intent inference and facial trait judgement. These analyses show how news media leverage images to visually persuade their audiences, and demonstrate the social dimensions of faces and their impact on real-world events.Finally, we will discuss the next steps for the project, including our newly-launched effort to begin digitizing local and national television news content dating back to 1970s. This is a joint work since 2010 between communications studies (Tim Groeling, Francis Steen) and Statistics/Computer Science (Weixin Li, Tao Yuan, Jungseock Joo, and Song-Chun Zhu), sponsored by the NSF CDI program CNS 1028381.
Event Details:
Parking | Directions
Please register here:
