DeepContext: Mobile Context Modeling and Prediction Via HMMs and Deep Learning

Abstract

Mobile context determination is an important step for many context-aware services such as location-based services, enterprise policy enforcement, building/room occupancy detection for power/HVAC operation, etc. Especially in enterprise scenarios where policies (e.g., attending a confidential meeting only when the user is in &#x201C;Location X&#x201D;) are defined based on mobile context, it is paramount to verify the accuracy of the mobile context. Most of the existing solutions rely on context obtained directly from sensors which could be hacked, noisy or insufficient, which cannot be relied upon for security applications. In this article, we take a different approach by modeling mobile context based on past context data of related users and considering its unique challenges such as missing features. To this end, we propose three models for modeling mobile context based on symbolic time series data of feature-value pairs&#x2014;two stochastic models based on the theory of Hidden Markov Models&#x00A0;(HMMs) and one model based on deep learning&#x2014;<i>personalized model</i> (<i>HPContext</i>), <i>collaborative filtering model</i> (<i>(HCFContext)</i>), and <i>deep learning model</i> (<i>DeepContext</i>) to eventually replace <i>HPContext</i>. <i>HPContext</i> and <i>DeepContext</i> predict the current context using sequential history of the user&#x0027;s own past context observations; while <i>HCFContext</i> enhances the former with collaborative filtering features, which enables it to predict the current context of the primary user based on the context observations of users related to the primary user, e.g., same team colleagues in the company, gym friends, family members, etc. <i>DeepContext</i> models mobile context based on symbolic (i.e., categorical valued rather than continuous-valued) time series data using deep learning techniques. These models are then used to determine the context of the primary user at the current instant or some timesteps in to the future. Each of the proposed models can also be used to enhance or complement the context obtained from sensors. Finally, these models are thoroughly validated on a real-life dataset.