Computing the Value of Spatio-Temporal Data in Wholesale and Retail Data Marketplaces

Spatio-temporal information is increasingly used for driving a plethora of intelligent transportation, smart-city, and crowd-sensing applications. At the same time, different types of data marketplaces are proposed for de-siloing and monetising individual and enterprise data. In this paper we study the problem of estimating the relative value of spatio-temporal data sold in wholesale and retail data marketplaces for the purpose of forecasting future demand in a certain area, e.g. a city. Using as case studies large datasets of taxi rides from Chicago and New York, we ask questions such as "When does it make sense for different taxi companies to combine their data?" and "How should different companies be compensated for the data that they share?". We then turn our attention to the even harder problem of establishing the value of the data brought to retail marketplaces by individual drivers. Overall, we show that simplistic approaches, such as assuming that the value of the data held by companies or drivers is proportional to its volume are inaccurate, because they fail to consider the complex complementarities that may exist among different datasets. To remedy this, more complex notions of value-sharing from economics and game-theory, such as the Shapley value need to be used to capture the effect of mixing datasets on the accuracy of forecasting algorithms driven by them. Applying the Shapley value to large datasets from many sources is computationally challenging. We use structured sampling to overcome such scalability challenges and manage to compute accurately the importance of different data sources, even when their number ranges in the thousands, as in the case of all the taxi drivers in a large metropolis.