From Geolocation-Based Only To Semantically-Aware Digital Advertising: A Neural Embedding Approach

The increasing availability of geo-location data in digital advertising businesses, like Valassis Digital, provides a vast amount of trajectories for mobile devices representing users location histories. As part of a unique massive data collection and analysis pipeline, a mapping stage from the geo-location data of mobile devices coming from advertisement exchange platforms to meaningful location atoms, such as geo-hashes, is included. Currently, location atoms are merely pieces of land in which their geographical properties connect them. However, online activities of mobile devices over time would impose an additional layer to the location data, which we call Semantical Location Layer (SLL). In fact, users online activities observed in different locations connect the location atoms not only by their geographical properties but also based on their sequences and co-occurrences. In this paper, we describe the developed SLL using a neural embedding algorithm to learn a latent representation of location atoms. Following this semantical layer, we develop a scalable collaborative filtering method to predict plausible location atoms for the devices. We show the results of the proposed method on very large geo-location data with a training set of size 200 million instances, and illustrate the impact of the proposed solution on the advertising business.