Multi-context embedding based personalized place semantics recognition

Abstract Personalized place semantics recognition is the process of giving individual semantic labels to locations, e.g., “home” and “school”. Capturing personalized place semantics exactly is critical for location-based services. To address the problems of existing methods, i.e., the insufficient utilization of context information and the neglect of the semantic correlation across related tasks, we propose a method for personalized place semantics recognition, which employs embedding methods, including deep learning based embedding and word embedding, to obtain effective representations from multi-context information (e.g., system settings, phone usage, and user activities). Meanwhile, we jointly model personalized place semantics and App usage sequences by sharing the App representations, which can improve generalization capability by exploiting the commonalities and differences across related tasks. We evaluate the proposed method on the Mobile Data Challenge dataset, and experimental results show that it outperforms existing methods significantly.