onsdag 7 december 2016

From UNESCO’s descriptive statistics to deductive Big Data: the role of human annotation in quantification processes

From IEDES, UMR 201, Université Paris 1 Panthéon-Sorbonne  
Analysis of the characteristics and activities of UNESCO’s statistical personnel indicates: (i) the intertwined nature of practices involved in data-production processes and of the knowledge this requires; (ii) the importance of human-annotation to production and maintenance of databases; (iii) the shift from descriptive statistics to statistical inference, all in the context of structured data. These findings help to define four recent quantification trends. As massive unstructured data takes over: (i) there is a greater reliance on machine learning and modeling; (ii) the use of supervised learning implies increasingly complex and diverse data annotation–which may modify the roleplayed by the social sciences; (iii) in unsupervised learning, based on non-annotated data, the role of statistical models is enhanced; (iv) in both cases inductive and deductive approaches may be of use. These trends are taken here to be represented by the expression “deductive quantification”.