Computational social scientists working with text often work with complex corpora that requires contextual understanding to be correctly classified and interpreted. Despite advances with Natural Language Processing (NLP) techniques, the approaches widely used today continue to struggle with understanding context. Recent advances in deep learning methods for supervised text classification help solve this problem. In this talk we describe Transformers models and their application to questions relevant to social scientists. These models improve contextual understanding of text and yield substantial improvements over current deep learning alternatives. We provide a set of three applications in the social sciences that demonstrate these advantages and show how researchers can apply them efficiently in their own work.
About the Speaker
Sebastian Vallejo Vera is an assistant professor at the School of Social Science and Government at the Tecnológico de Monterrey, México. He is the director of the interdisciplinary Laboratory of Computational Social Science - México (iLCSS). His research explores the relation between gendered political institutions and representation, and racial identity and racism in Latin America. Vallejo Vera's methodological work applies novel Natural Language Processing (NLP) to a wide variety of text data, from legislative speeches to tweets, to answer substantive questions about gender, racism, and politics. You can find more information at www.svallejovera.com.
