Collection of Twitter Corpora for Human and Social Sciences: Sampling Methodology and Evaluation

Abstract

The increasing popularity of electronic messages challenges social science researchers, particularly regarding data representativeness and quality. We propose and evaluate a methodology to create a corpus of Twitter data for a given population by sampling the targeted user population. In our case, the population is the groups of citizens, media and politicians using Twitter in Belgium (French and Dutch languages), Norway and France. We present in particular a machine-learning based methodology that enables the population sampling. We also present a methodology to evaluate the representativeness of our corpus compared to the Full Twitter stream of the targeted population.

Topics & Concepts