Social-minded Measures of Data Quality

Abstract

For decades, research in data-driven algorithmic systems has focused on improving efficiency (making data access faster and lighter) and effectiveness (providing relevant results to users). As data-driven decision making becomes prevalent, there is an increasing need for new measures for evaluating the quality of data systems. In this article, we make the case for social-minded measures, that is, measures that evaluate the effect of a system in society. We focus on three such measures, namely diversity (ensuring that all relevant aspects are represented), lack of bias (processing data without unjustifiable concentration on a particular side), and fairness (non discriminating treatment of data and people).