We show that the word embedding technique word2vec is mathematically equivalent to the gravity law of mobility, making it ideal for learning dense representations from migration data that can be distributed, re-used, and studied. By treating locations analogously to words and trajectories to sentences,we demonstrate the power of word2vec by applying it to the case of scientists’ migrations, for which it encodes information about culture, geography, and prestige at multiple layers of granularity. Our results lay a theoretical and methodological foundation for the application of neural embeddings to the study of migration.
Dakota Murray, Jisung Yoon, Sadamori Kojaku, Rodrigo Costas, Woo-Sung Jung, Staša Milojević, Yong-Yeol Ahn