Learning human mobility behaviors from location-sensing data are crucial to mobility data mining because of its potential to address a range of analytical purposes in mobile context reasoning, including exploration, inference, and prediction. However, existing approaches suffer from two practical problems: temporal and spatial sparsity. To address these shortcomings, we present two unsupervised learning methods to model the mobility behaviors of multiple users (i.e., a population), considering efficiency and accuracy. These methods intelligently overcome the sparsity in individual data by seeking temporal commonality among users’ heterogeneous location behaviors. The advantages of our models are highlighted through experiments on several real-world mobility data sets, which also show how our methods can realize the three analytical purposes in a unified manner.

Inverse reinforcement learning (IRL) aims to recover the reward function underlying a Markov Decision Process from behaviors of experts in support of decision-making. Most recent work on IRL assumes the same level of trustworthiness of all expert behaviors, and frames IRL as a process of seeking reward function that makes those behaviors appear (near)-optimal. However, it is common in reality that noisy expert behaviors disobeying the optimal policy exist, which may degrade the IRL performance significantly. To address this issue, in this paper, we develop a robust IRL framework that can accurately estimate the reward function in the presence of behavior noise. In particular, we focus on a special type of behavior noise referred to as sparse noise due to its wide popularity in real-world behavior data. To model such noise, we introduce a novel latent variable characterizing the reliability of each expert action and use Laplace distribution as its prior. We then devise an EM algorithm with a novel variational inference procedure in the E-step, which can automatically identify and remove behavior noise in reward learning. Experiments on both synthetic data and real vehicle routing data with noticeable behavior noise show significant improvement of our method over previous approaches in learning accuracy, and also show its power in de-noising behavior data.

Far beyond relationship topology, today's online social networks are also characterized by semantically rich text messages exchanged among users as well as GPS locations associated with those messages, as evidenced by Twitter's geotagged tweets. Textual contents help characterize users' personal interests, while geographical features help link users' behaviors in the online world to those in the physical world such as their mobility patterns. In this paper, instead of studying each aspect separately, as done by most previous works, we combine textual contents and spatial features in a joint way using Bayesian latent topic model in order to construct better algorithms for user characterization and social network study. Specifically, the integration of contents and spatial features in a user-centered environment can not only discover geographic topics but also enable the characterization of users' latent interests with geographic semantics. Such a novel characterization can be leveraged to benefit many interesting studies regarding social network heterogeneity and relationships between online networks and physical world. Using a large-scale twitter data set with broad geographical coverage, we systematically evaluate our framework in several typical inference tasks surrounding user, content and location, as well as carry out empirical studies in real world scenarios. Experimental results demonstrate the advantages of our joint modeling approach, as well as its potentials to facilitate user understanding, both in online world and physical world.

Road latent cost, which quantifies how desirable each road is for traveling, is important information to enable many smartcity applications such as route recommendation. Arguably, vehicle trajectories are a good source to learn these costs as drivers intelligently incorporate them into their routing decisions. However, major past approaches misinterpret drivers' behaviors and suffer from trajectory sparsity problem, mainly because they adopt an edge-centric perspective which fails to exploit the sequential information in the entire trajectories. To address these shortcomings, we model drivers' routing decision process which targets at global path optimality, and present a framework to reliably discover those costs by exploiting entire trajectories while isolating the influence of heterogeneous destinations. Extensions are also made to address several issues in practice. Extensive experiments on real world data show that the road costs learned in this way significantly outperform past approaches in several urban computing tasks and require less data for learning.

The extraction of macroscopic mobile context reflecting users' personal and social behavior patterns from smartphone sensor data (e.g., GPS/Bluetooth signals) is crucial in building intelligent pervasive systems. Hierarchical Dirichlet Processes (HDP), a well known Bayesian nonparametrics model for grouped data, is a promising option to achieve this objective due to its ability of discovering high-level semantics behind raw signals and establishing connections between individuals. However, applying HDP in a straightforward manner may not work as it does not take certain unique characteristics in mobile context into account. Particularly, while traditional HDP typically models a single aspect (e.g., Word), the characterization of a mobile context normally involves multiple heterogeneous aspects (e.g., Time, location, Bluetooth proximity). In addition, the presence of multiple aspects dictates a flexible way of clustering users and organizing mobile contexts in a hierarchical manner in serving different pervasive applications, a feature that traditional HDP lacks. Therefore, in this paper, we propose several extensions on traditional HDP to adapt it to the task of mobile context discovery. The key features in our extensions are: i) fusing multiple aspects naturally in HDP to achieve effective extraction of complex mobile context, ii) treating different aspects heterogeneously (globally or personally) in HDP to enable flexible user behavior clustering at various granularities in accordance with applications' needs, and iii) organizing mobile contexts in a hierarchical manner for natural behavior representation and overcoming data sparsity. Based on the experiments in a popular real-world mobile data set, we illustrate the ability of the framework in extracting useful mobile contexts such as characterizing personal life routines, discovering dominant temporal habits in a population, and inferring social group patterns, as well as its potential in improving individual mobility prediction under data sparsity.

Road travel costs are important knowledge hidden in large-scale GPS trajectory data sets, the discovery of which can benefit many applications such as intelligent route planning and automatic driving navigation. While there are previous studies which tackled this task by modeling it as a regression problem with spatial smoothness taken into account, they unreasonably assumed that the latent cost of each road remains unchanged over time. Other works on route planning and recommendation that have considered temporal factors simply assumed that the temporal dynamics be known in advance as a parametric function over time, which is not faithful to reality. To overcome these limitations, in this paper, we propose an extension to a previous static trajectory regression framework by learning the temporal dynamics of road travel costs in an innovative non-parametric manner which can effectively overcome the temporal sparsity problem. In particular, we unify multiple different trajectory regression problems in a multi-task framework by introducing a novel cross-task regularization which encourages temporal smoothness on the change of road travel costs. We then propose an efficient block coordinate descent method to solve the resulting problem by exploiting its separable structures and prove its convergence to global optimum. Experiments conducted on both synthetic and real data sets demonstrate the effectiveness of our method and its improved accuracy on travel time prediction.

In recent years, researchers have sought to capture the daily life location behaviour of groups of people for exploratory, inference, and predictive purposes. However, development of such approaches has been limited by the requirement of personal semantic labels for locations or social/spatial overlap between individuals in the group. To address this shortcoming, we present a Bayesian model of mobility in populations (i.e., groups without spatial or social interconnections) that is not subject to any of these requirements. The model intelligently shares temporal parameters between people, but keeps the spatial parameters specific to individuals. To illustrate the advantages of population modelling, we apply our model to the difficult problem of overcoming data sparsity in location prediction systems, using the Nokia dataset comprising 38 individuals, and find a factor of 2.4 improvement in location prediction performance against a state-of-the-art model when training on only 20 hours of observations.

Understanding a user's social interactions in the physical world proves important in building context-aware ubiquitous applications. A good way towards that objective is to categorize people to whom a user is socially related into what we call as social circles. In this note, we propose a novel unsupervised approach that learns from the Bluetooth (BT) sensed data recording one's dynamic proximity relations with others to identify her social circles, each of which is formed along a semantically coherent aspect. For each circle we learn its members as well as the temporal dimensions along which it is formed. Our method is innovative in that it well overcomes data sparsity by information sharing, and allows for circle overlaps which is common in reality. Experiments on real data demonstrate the effectiveness of our method, and also show the potentials of relational mobile data in sensing personal behaviors beyond personal data.

Recognizing and classifying users' routine behavior patterns from sensor data has been a hot topic in pervasive computing. Its objective is to automatically discover recurrent routine patterns in a user's daily life by leveraging the multimodal data generated from wearable sensors such as mobile phones. This kind of knowledge can be utilized in many ways such as identifying similar users in terms of their behaviors, providing behavior contexts to enable advanced human-centered applications, etc. While numerous works have been done in this area, most of them rely on densely sampled mobile data collected from specially-programmed sensors that can “follow” people throughout the day. In this paper, we study how to achieve the same objective when the mobile data presented is much sparser, such as traditional mobile phone data where a user's location is reported only when he makes a call. Although a single user's mobile data is far from sufficient to reveal his characteristic behavior, we show that when exploiting a large number of users' mobile data in a principled collaborative way which facilitate similar users' data to complement each other, representative routine patterns can be revealed and each user can be characterized properly. Experiments on synthetic and real mobile phone data set demonstrate the effectiveness of our methods, and also show our model's ability in predicting human activity using the patterns learned.

Human behavior understanding is a fundamental problem in many ubiquitous applications. It aims to automatically uncover and quantify characteristic behavior patterns in users' daily lives as well as disclose behavior clustering structure among multiple users. The key challenge is how to define a naturally interpreted representation for users' daily behavior patterns, which can be easily exploited to not only uncover the behavior similarity among multiple users but also predict users' future activities. In this paper, we define such a representation, and propose a probabilistic framework which can automatically learn it from mass amount of mobile data in unsupervised setting and exploit it to predict user activities. By an appropriate information sharing among multiple users, this framework overcomes single-user data sparsity problem and effectively identifies behavior clustering structures in a set of users. Experiments conducted on a public reality mining data set demonstrate the effectiveness and accuracy of our methods.