User Location Prediction using MLPsMotivationPast ResearchDatasetPerceptron focus: next nodeConclusions/Further StudyQuestions?A STUDY ON THE USE OF MULTILAYER PERCEPTRONS TO PREDICT FUTURE LOCATIONS BASED ON PAST LOCATION AND TIME INFORMATIONUser Location Prediction using MLPsHans WegmuellerMotivationMy design project – an application that pushes location information to existing social networks, and leverages those social networks to share location information from contacts.Understanding user location prediction, its limitations and potential, could lead to further services being provided to users, and provide valuable information to advertisers.Past Research“Mobile User Movement Prediction Using Bayesian Learning for Neural Networks”Sherif Akoush, American University in Cairo“A Predictive Location Model for Location-Based Services”Hassan A. Karimir and Xiong Liu, University of PittsburghDatasetGenerated dataset includes location “nodes,” day’s of the week, and time of locations.Assumes that peoples movements day-to-day follow a pattern, but that the clarity of that pattern differs on a person to person basis.Dataset generated falls into 4 ‘types’ of people, ranging from someone totally regimented to someone who moves randomly between N nodes each day.Perceptron focus: next nodeBuild a 3 layer perceptron:Day of the 14 day cycle maskTime period maskWeight most visited locations during day/time periodFoundAs one might expect, the more regimented the pattern, the more easy it is to predict next location.Conclusions/Further StudyPerceptron design likely a function of dataset, only real data could determine if the assumptions made during dataset generation yields a useful model for location predictionWould like to further study designing for variance of timeThen would like to have multiple MLPs that can be chosen between depending on the user, and their past accuracyQuestions?Thank
View Full Document