Predictive Location-Based Services Foundationally
Location Sensor User Models a model is a collection of data on some particular aspect of a human user's behaviour that, when associated with a limited set of contextual clues, yields predictions on what behavious the human will engage in next. Concepts E-911 system Not very predictive Minimizing cell-phone system infrastructure for finding you. Hypothetical system Predictive model could help a lot Active Badge System Properties of this system Example of how to build a predictive system Markov-Model based solution GPS requires a way of determining significant places Predictive Location Based Services with GPS Shopping List Annoying to be told when you are driving by the store that you have things you need to buy If you knew a few blocks in advance, not so annoying Library If a system knew you were going by a library later in the day You could be prompted to bring along a book If you were wearing a wearable computer You could be optimize queued email sending You could be alerted to go a different direction to get infrastructure You could hold the message even longer for power concerns Multi-User Applications Combine two user models to answer the question Will I see Ryan today? How can we answer this while maintaining Ryan's privacy? Favor arbitrage Model-sharing Share partial models Store models in a central server cache coherency issues Query when needed Using GPS to learn significant locations and predict movement across multiple users Experiment Study One one user four months recorded position at one second intervals Look at data Finding Significant Locations The idea is to predict transitions between significant locations Significant locations are places where you spend time Is this a good definition? What's a better definition? How much time Look at data Clustering places Look at clustering algorithm Figuring out radius Look at knee Markov Model on these locations Considering the order of the Markov Model Look at data Study Two repeat in Zurich with 6 people tried to study semantic places Future Work Time prediction