Mining Users' Behaviors and Environments for Semantic Place Prediction

work, we propose a novel prediction framework, which takes into account the spatial property, temporal property, users' behavior and environment at the same time, for semantic place prediction. The core idea of our proposal is to extract features to represent end users' behaviors in each place related to its semantic. To achieve this goal, we define 54 features to represent end users' behaviors to capture the key properties of places recorded in MDC Data Set. In our framework, we propose a novel model, namely Multi-Level Classification Model, to solve the imbalanced data problem. Based on the Multi-Level Classification Model, we make semantic prediction of a place by combining several classification models. To our best knowledge, this is the first work on predicting semantic label of places through integrating sub-classification models into a multi-level structure.