Hybrid Format

Implications of assigning nationality to names using machine learning on the scholarly migration rates worldwide

Faeze Ghorbanpour
Department of Digital and Computational Demography, August 30, 2023

Hybrid Seminar Talk, August 30 from 11am to 12pm (CEST)
 

Abstract

Different from register data, observational and digital trace data usually do not include the nationality of individuals. Detecting the nationality of people plays a key role in developing fair and privacy-aware machine learning models. It enables answering important questions in demography and social sciences. Our aim was to detect the nationality with the least available data, people’s full names. We gathered 3 million name-nationality pairs from Wikipedia. We applied char-based machine learning models to categorize these names into families of nationalities such as German, Spanish, and English, to name a few. This categorization has three granularity levels from more general geographical regions to country regions. Our model surpasses the performance of the models discussed in the literature, demonstrating a 10% enhancement in terms of the weighted f1-score. To analyze migration rates among Scopus authors on a global scale, we utilized assigned nationalities. The outcomes of our analysis reveal that this approach provides valuable insights into the shifts in migration rates when considering the country of origin based on authors’ names rather than previously used academic affiliations as the country of “academic birth”.

About the speaker

Faeze is a Guest Researcher at the Migration and Mobility laboratory of the Department of Digital and Computational Demography at the MPIDR, Rostock; She has a Master's degree in Computer Engineering from the Sharif University of Technology, Tehran, Iran. She is an upcoming PhD student at the Center of Information and Language Processing at the Ludwig Maximilian University of Munich.