Unsupervised embedding of trajectories captures the latent structure of scientific migration

Abstract

Human migration and mobility drives major societal phenomena including epidemics, economies, innovation, and the diffusion of ideas. Although human mobility and migration have been heavily constrained by geographic distance throughout the history, advances, and globalization are making other factors such as language and culture increasingly more important. Advances in neural embedding models, originally designed for natural language, provide an opportunity to tame this complexity and open new avenues for the study of migration. Here, we demonstrate the ability of the model word2vec to encode nuanced relationships between discrete locations from migration trajectories, producing an accurate, dense, continuous, and meaningful vector-space representation. The resulting representation provides a functional distance between locations, as well as a “digital double” that can be distributed, re-used, and itself interrogated to understand the many dimensions of migration. We show that the unique power of word2vec to encode migration patterns stems from its mathematical equivalence with the gravity model of mobility. Focusing on the case of scientific migration, we apply word2vec to a database of three million migration trajectories of scientists derived from the affiliations listed on their publication records. Using techniques that leverage its semantic structure, we demonstrate that embeddings can learn the rich structure that underpins scientific migration, such as cultural, linguistic, and prestige relationships at multiple levels of granularity. Our results provide a theoretical foundation and methodological framework for using neural embeddings to represent and understand migration both within and beyond science.

Understanding scholar-trajectories across scientific periodicals

“…Figure 2 c–d plots some examples of author mobility in the embedding space—as a researcher's career progresses, her publishing venue moves between different periodicals (i.e., journals and proceedings), forming a unique publication trajectory in the space of scientific periodicals. Prior studies have demonstrated that the embedding models can learn a systematic representation of mobility in the cases of institution locations 16 and research topics 15 . Hence, we expect the reconstructed trajectories from the embeddings of scientific periodicals to provide meaningful information about individual scientific activities and career profiles.…”

“…To reduce the complexity of trajectory data, we use single quantitative measures. Here, we introduce four measures, including the mean embedding distance 16 , the average distance to the midpoint, the radius of gyration in the original space, and the radius of gyration in the 2-d space 23 . In particular, the radius of gyration, , is a well-known indicator in mobility analysis that measures the characteristic distance traveled by an individual 27 – 30 .…”

“…To characterize the trajectory patterns, we extend the human mobility measure, the radius of gyration, from the geographic realm to the geometric realm 15 , 23 . Many other measures can be studied concerning individual trajectory patterns, such as the embedding-based distance 16 , the Shannon Entropy 51 , and the convex hull area 52 (See Fig. S12 for the correlation between an example set of measures).…”

“…We calculate the distance between locations of periodicals d ij based on the cosine similarity, where and are the embeddings for periodicals i and j , respectively. In a recent study of trajectory embeddings 16 , this cosine similarity-based embedding distance outperforms other forms of distance measures. Likewise, we estimate the average distance between individual publication positions and the center of the trajectory, , using this cosine similarity-based embedding distance, where N is the total number of publications, K is the set of periodicals visited by an individual, is the number of visits of the periodical i , cm is the trajectory's center of mass, i.e., the weighted mean of the periodicals visited by the author.…”