Fairness, Accountability, and Transparency while Mining Data from the Web and Social Networks

Abstract

Digital media have been changing fundamentally our society, as a consequence of easier access to contents as well as better and cheaper generation and dissemination through the internet, as witnessed by services such as online videos, games and social networks. More recently, there has been an increasing availability of "smart" services that, among other tasks, help users to locate, understand and analyze automatically media of interest. Smart services are often based on algorithms from data mining and related areas such as machine learning and artificial intelligence. Beyond the efficiency and effectiveness of theses services, there is a growing concern about the fairness, accountability and transparency associated with them, which is the subject of this talk. Fairness comprises guarantees that algorithms are neither biased nor discriminatory, even when they are mathematically and computationally correct. Accountability means the identification of entities, human or not, that should be held responsible for the algorithms' consequences. Transparency is the property of generating understandable explanations on the algorithms' outcomes. In this talk we are going to discuss and characterize data mining algorithms, in particular when applied to web and social networks, with respect to fairness, accountability and transparency, and present strategies that assure these properties while satisfying other usual requirements such as precision, effectiveness, and privacy preservation.