PROJECT TITLE :

Systematic Analysis of Fine-Grained Mobility Prediction with On-Device Contextual Data

ABSTRACT:

The concept of predicting the mobility of users is widely discussed within the research community. Numerous studies have investigated a wide variety of algorithms with the goal of determining, based on the contexts and trajectories of users, the locations that users are most likely to visit. The majority of the studies that are currently available center on particular predictions' targets. Although successful cases are frequently reported, there have been relatively few discussions on what occurs if the prediction targets vary. For example, it has not been determined whether coarser locations are easier to predict or whether predicting the next location on the trajectory immediately after the current one is simpler than predicting the destination. On the other hand, while spatiotemporal tags and content information are commonly used in current prediction tasks, few have utilized the finer grained, on-device user behavioral data, which are supposed to be more informative and indicative of user intentions. This is because on-device user behavior data are collected locally on the user's device. In this paper, we conduct a methodical study on the prediction of mobility using a large-scale real-world dataset that includes a wealth of contextual information. Extensive experiments are carried out by us on the basis of a number of learning models, among which are a Markov model, two models of recurrent neural networks, and a method of multi-modal learning. The purpose of these experiments is to investigate in depth the predictability of various types of granularities of targets as well as the efficiency of various types of signals. The findings offer illuminating information on what can be predicted along with how it can be done, which sheds light on the real-world mobility prediction from a more general point of view.