Although efforts to improve the accountability of social media algorithms have been made in recent years, many have also expressed skepticism about data collection approaches that rely on ‘corporate data philanthropy.’ As an alternative to industry-academic partnership approaches, our research team proposed data donation from research participants as a route to audit algorithms. Drawing on our experience with three data donation collections, I argue that not only can the donation approach provide researchers and civic society with reachable data access, but also it is expected to afford stable one as platform users’ access rights to their own data are increasingly required by the consumer protection laws. Yet, the new approach also poses methodological challenges, such as ‘small N, large P’ problem and selection bias mediated by privacy concerns and exceeding participants’ technical capacities. I will discuss potential innovations to tackle these challenges, such as data augmenting and sample weighting.
Chankyung Pak is Assistant Professor in Division of Business and Management (DBM) at Beijing Normal University-Hong Kong Baptist University (BNU-HKBU UIC). He is a computational social scientist who’s interested in the evolution of public discourse mediated by transforming media ecosystem. In most of his works, he tries to put media studies, information science and economics together. Currently, he works with Dr. Magdalena Wojcieszak for an ERC project on backfire effects that follow exposure to dissimilar views online. While he is an avid learner of economic theory, mathematical argument and machine learning, he started as a cultural study scholar and still is looking to connect his research back with critical theories on algorithms and artificial intelligence. He recently defended his dissertation, “News Organizations’ News Link Sharing Strategies on Twitter: Economic Theory and Computational Text Analysis,” in the Department of Media and Information at Michigan State University. He is also a former member of Behavior, Information and Technology Lab (BITLab).