The spread of an infectious disease through a population can be modeled using a network or a graph. In digital advertising, internet device graphs are graph data sets that organize identifiers produced by mobile phones, PCs, TVs, and tablets as they access media on the internet. Characterized by immense scale, they have become ubiquitous as they enable targeted advertising, content customization and tracking. This paper posits that internet device graphs, in particular those based on IP colocation, can provide significant utility in predicting and modeling the spread of infectious disease. Starting the week of March 16th, 2020, in the United States, many individuals began to `shelter-in-place as schools and workplaces across the nation closed because of the COVID-19 pandemic. This paper quantifies the effect of the shelter-in-place orders on a large scale internet device graph with more than a billion nodes by studying the graph before and after orders went into effect. The effects are clearly visible. The structure of the graph suggests behavior least conducive to transmission of infection occurred in the US between April 12th and 19th, 2020. This paper also discusses the utility of device graphs for i) contact tracing, ii) prediction of `hot spots, iii) simulation of infectious disease spread, and iv) delivery of advertisement-based warnings to potentially exposed individuals. The paper also posits an overarching question: can systems and datasets amassed by entities in the digital ad ecosystem aid in the fight against COVID-19?
Download