Handling location uncertainty in event driven experimentation

Abstract

The wide spread use of smart phones has ushered in a wave of context-based advertising services that operate on pre-defined user events. A prime example is Location Based Advertising. What is missing though, is the ability to experiment with these services under varying event conditions with real users using their regular phones in real-world environments. Such experiments provide greater insight into user needs for and responsiveness towards context-based advertising applications. However, these event-driven experiments rely on data that arrive from sources such as mobile sensors which have inherent uncertainties associated with them. This effects the interpretation of the outcome of an experiment. In this paper we introduce Jarvis, a behavioural experimentation platform that supports running in-situ real-time experiments of mobile advertising services, targeting real participants on their smart phones based on multiple context-specific events. We highlight the challenges of handling uncertainty on such a platform as well as how we deal with ambiguity in the location attribute.