On the 28th of August, we visited sunny Georgetown University to discuss all things politics and Computational Social Science for the second annual PaCSS conference. For the last two years, SAGE Ocean has been proud to sponsor PaCSS, which aims to bring together scholars across politics, computer science, and related fields studying political communication and behavior occurring online. The drivers for launching PaCSS really resonate with the reasons we set up SAGE Ocean, as outlined in our mission statement.
“The data and methodologies available to social scientists have exploded with the emergence of vast archives of passive data collection, large scale online experimentation, and innovative uses of simulation. These data are of a larger magnitude and methods are of a greater computational complexity than approaches that have dominated political science for the last 50 years. This offers the potential for rich insights into society at scale, while simultaneously introducing new ethical and infrastructural challenges.”
The focus on image and video
Online image analysis feels like it could be the next big thing. As Jeffrey Sternberg from Northeastern University said during his presentation, Using Computer Vision to Capture the Collective Perception of a Neighborhood, we are at the point where we now have the technology to be able to treat images as data in much the same way we treat text and using machine learning we can do classification, clustering and information and pattern extraction. The question for the social sciences is, what do we want to use image analysis for? What kinds of social science questions can it help us to answer, and what kinds of analysis methods are suited to image analysis? Across the two streams on images and video, we got a nice set of examples of social science questions visual data can help us to answer, including whether women candidates “run as women” online from Jielu Yao, Wesleyan University & University of Iowa, and mapping extremist networks online using visual images from propaganda videos from Rob Williams at UNC Chapel Hill.
One of my favourite talks of the conference was from Emma Remy from the Pew Research Center: How do Machines See Gender? Demystifying a machine vision system. The aim of their research was to “look under the hood” of an algorithm they trained to classified individuals who appeared in Google images as male or female. What they found was pretty surprising (but they haven’t published their findings yet, so I don’t want to steal their thunder). They have published a great blog post talking about how they used transfer learning to make their classification project more manageable though, which I’d recommend to social scientists starting out with deep learning.
The variety of social media research
As you’d expect from this conference, there were plenty of talks centered around social media data analysis; Yotam Smargard from the University of Arizona presented a talk on The Influencer Ecosystem in the 2018 US Primaries; Dartmouth College’s Jin Woo Kim spoke about The Distorting Prism of Social Media: How Online Comments Amplify Toxicity; and Josh Tucker from NYU presented his research Analyzing Link Sharing Across Platforms to Study Political Messaging and Ideology.
However, we also saw some interesting examples of how social media can be used to run experiments and reach participants in new ways. Leah Rosenzweig from the Institute for Advanced Study in Toulouse (IAST) gave a talk on Social Media Markets for Survey Research in Comparative Contexts: Facebook Users in Kenya. Her research project had been designed to find out if Facebook could offer a cheaper, faster and reliable mechanism to recruit online survey respondents in developing countries. The pilot project helped her to understand how online survey respondents recruited through Facebook ads compared against offline samples and has laid the groundwork for further research.
The excellent keynote
Sandra González-Bailón’s keynote on Exposure to News in the Digital Age walked through the problems in measuring online news exposure, the biases included in measuring online news exposure and the long tail of news online. She also shared details of her fascinating current project looking at Digital News and the Consumption of Information Online which is funded by the NSF.
PaCSS was a highlight in our calendar last year and this year’s event did not disappoint! It was a great crowd, a fab keynote, and a bunch of fascinating computational social science papers on a diverse range of topics. Check out what other attendees had to say about the event on Twitter.
Thanks to all the organizing committee - we look forward to hearing where PaCSS 2020 will be held. Watch this space!