Hongjian Wang

Compressive sensing based monitoring with vehicular networks

Vehicles are becoming powerful mobile sensors, and vehicular networks provide a promising platform to support a wide range of existing large-scale monitoring applications such as road surface monitoring, and etc. In vehicular networks, inter-vehicle contacts are scarce resources for data delivery. This presents a major challenge for monitoring applications with vehicular networks. By analyzing a large dataset of taxi traces collected from around 2,600 taxis in Shanghai, China, we reveal that there is strong correlation with data readings on vehicles. Motivated by this important observation, we propose a compressive sensing based approach called CSM to monitor with vehicular networks. Two key issues must be addressed. First, there is an intrinsic tradeoff between communication cost and estimation accuracy. Second, guaranteed estimation accuracy should be provided over the highly dynamic network. To address the above issues, we first characterize the relationship between estimation error (12 error) and sparsity property of a dataset. Then, we determine two critical parameters: the minimum number of seeds and the minimum transmission hop length for compressive measurements in the network. The selection of the two parameters can reduce the communication cost while guaranteeing the required estimation accuracy. Extensive simulations based on real vehicular GPS traces collected in Shanghai, China have been performed and results demonstrate that CSM achieves much higher estimation accuracy at the same communication cost compared with other alternative schemes.

Semantic Annotation of Mobility Data using Social Media

Recent developments in sensors, GPS and smart phones have provided us with a large amount of mobility data. At the same time, large-scale crowd-generated social media data, such as geo-tagged tweets, provide rich semantic information about locations and events. Combining the mobility data and surrounding social media data enables us to semantically understand why a person travels to a location at a particular time (e.g., attending a local event or visiting a point of interest). Previous research on mobility data mining has been mainly focused on mining patterns using only the mobility data. In this paper, we study the problem of using social media to annotate mobility data. As social media data is often noisy, the key research problem lies in using the right model to retrieve only the relevant words with respect to a mobility record. We propose frequency-based method, Gaussian mixture model, and kernel density estimation (KDE) to tackle this problem. We show that KDE is the most suitable model as it captures the locality of word distribution very well. We test our proposal using the real dataset collected from Twitter and demonstrate the effectiveness of our techniques via both interesting case studies and a comprehensive evaluation.

PGT: Measuring Mobility Relationship Using Personal, Global and Temporal Factors

Rich location data of mobile users collected from smart phones and location-based social networking services enable us to measure the mobility relationship strength based on their interactions in the physical world. A commonly-used measure for such relationship is the frequency of meeting events (i.e., Co-locate at the same time). That is, the more frequently two persons meet, the stronger their mobility relationship is. However, we argue that not all the meeting events are equally important in measuring the mobility relationship and propose to consider personal and global factors to differentiate meeting events. Personal factor models the probability for an individual user to visit a certain location, whereas the global factor models the popularity of a location based on the behavior of general public. In addition, we introduce the temporal factor to further consider the time gaps between meeting events. Accordingly, we propose a unified framework, called PGT, that considers personal, global, and temporal factors to measure the strength of the relationship between two given mobile users. Extensive experiments on real datasets validate our ideas and show that our method significantly outperforms the state-of-the-art methods.

A simple baseline for travel time estimation using large-scale trip data

The increased availability of large-scale trajectory data provides rich information for the study of urban dynamics. For example, New York City Taxi & Limousine Commission regularly releases source/destination information of taxi trips, where 173 million taxi trips released for Year 2013 [1]. Such a big dataset provides us potential new perspectives to address the traditional traffic problems. In this paper, we study the travel time estimation problem. Instead of following the traditional route-based travel time estimation, we propose to simply use a large amount of taxi trips without using the intermediate trajectory points to estimate the travel time between source and destination. Our experiments show very promising results. The proposed big data-driven approach significantly outperforms both state-of-the-art route-based method and online map services. Our study indicates that novel simple approaches could be empowered by the big data and these approaches could serve as new baselines for some traditional computational problems.

Region Representation Learning via Mobility Flow

Increasing amount of urban data are being accumulated and released to public; this enables us to study the urban dynamics and address urban issues such as crime, traffic, and quality of living. In this paper, we are interested in learning vector representations for regions using the large-scale taxi flow data. These representations could help us better measure the relationship strengths between regions, and the relationships can be used to better model the region properties. Different from existing studies, we propose to consider both temporal dynamics and multi-hop transitions in learning the region representations. We propose to jointly learn the representations from a flow graph and a spatial graph. Such a combined graph could simulate individual movements and also addresses the data sparsity issue. We demonstrate the effectiveness of our method using three different real datasets.

Interpreting traffic dynamics using ubiquitous urban data

Given a large collection of urban datasets, how can we find their hidden correlations? For example, New York City (NYC) provides open access to taxi data from year 2012 to 2015 with about half million taxi trips generated per day. In the meantime, we have a rich set of urban data in NYC including points-of-interest (POIs), geo-tagged tweets, weather, vehicle collisions, etc. Is it possible that these ubiquitous datasets can be used to explain the city traffic? Understanding the hidden correlation between external data and traffic data would allow us to answer many important questions in urban computing such as: If we observe a high traffic volume at Madison Square Garden (MSG) in NYC, is it because of the regular peak hour or a big event being held at MSG? If a disaster weather such as a hurricane or a snow storm hits the city, how would the traffic be affected? Most of existing studies on traffic dynamics focus only on traffic data itself and do not seek for external datasets to explain traffic. In this paper, we present our results in attempts to understand taxi traffic dynamics in NYC from multiple external data sources. We use four real-world ubiquitous urban datasets, including POIs, weather, geo-tagged tweets, and collision records. To address the heterogeneity of ubiquitous urban data, we present carefully-designed feature representations for these datasets. Our analysis suggests that POIs can well describe the regular traffic patterns. In addition, geo-tagged tweets can be used to explain irregular traffic caused by big events, and weather may account for abnormal traffic drops.

SemMobi: A Semantic Annotation System for Mobility Data

The wide adaptation of mobile devices embedded with modern positioning technology enables the collection of valuable mobility data from users. At the same time, the large-scale user-generated data from social media, such as geo-tagged tweets, provide rich semantic information about events and locations. The combination of the mobility data and social media data brings opportunities for us to study the semantics behind peoples movement, i.e., understand why a person travels to a location at a particular time. Previous work have used map or POI (point of interest) database as source for semantics. However, those semantics are static, and thus missing important dynamic event information. To provide dynamic semantic annotation, we propose to use contextual social media. More specifically, the semantics could be landmark information (e.g., a museum or an arena) or event information (e.g., sports games or concerts). The SemMobi system implements our recently developed annotation method, which has been recently accepted to WWW 2015 conference. The annotation method annotates words to each mobility records based on local density of words, estimated by Kernel Density Estimation model. The annotated mobility data contain rich and interpretable information, therefore can benefit applications, such as personalized recommendation, targeted advertisement, and movement prediction. Our system is built upon large-scale tweet datasets. A user-friendly interface is designed to support interactive exploration of the result.

Compressive data retrieval with tunable accuracy in vehicular sensor networks

On-demand data retrieval is a crucial routine operation in a vehicular sensor network. However, on-demand data retrieval in a vehicular environment is particularly challenging because of frequent network disruption, large number of data readings and limited transmission opportunities. Real world vehicular datasets usually contain a lot of data redundancy. Motivated by this important observation, we propose an approach called CDR with compressive sensing for on-demand data retrieval in the highly dynamic vehicular environment. The distinctive feature of CDR is that it supports tunable accuracy of data collection. There are two major challenges for the design of CDR. First, the sparsity level of the vehicular dataset is typically unknown beforehand. Second, it is even worse that the sparsity level of the dataset is changing over time. To combat the challenge posed by time-varying data sparsity, CDR can terminate from further collection of measurements, based on an adaptive condition on which only localized measurements and computation are needed. Extensive simulations with real datasets and real vehicular GPS traces show that our approach achieves good performance of data retrieval with user-customized accuracy.

Representation Learning for Large-Scale Dynamic Networks

Representation leaning on networks aims to embed networks into a low-dimensional vector space, which is useful in many tasks such as node classification, network clustering, link prediction and recommendation. In reality, most real-life networks constantly evolve over time with various kinds of changes to the network structure, e.g., creation and deletion of edges. However, existing network embedding methods learn the representation vectors for nodes in a static manner, which are not suitable for dynamic network embedding. In this paper, we propose a dynamic network embedding approach for large-scale networks. The method incrementally updates the embeddings by considering the changes of the network structures and is able to dynamically learn the embedding for networks with millions of nodes within a few seconds. Extensive experimental results on three real large-scale networks demonstrate the efficiency and effectiveness of our proposed methods.

Inferring Mobility Relationship via Graph Embedding

Inferring social relationships from user location data has become increasingly important for real-world applications, such as recommendation, advertisement targeting, and transportation scheduling. Most existing mobility relationship measures are based on pairwise meeting frequency, that it, the more frequently two users meet (i.e., co-locate at the same time), the more likely that they are friends. However, such frequency-based methods suffer greatly from data sparsity challenge. Due to data collection limitation and bias in the real world (e.g., check-in data), the observed meeting events between two users might be very few. On the other hand, existing methods focus too much on the interactions between two users, but fail to incorporate the whole social network structure. For example, the relationship propagation is not well utilized in existing methods. In this paper, we propose to construct a user graph based on their spatial-temporal interactions and employ graph embedding technique to learn user representations from such a graph. The similarity measure of such representations can well describe mobility relationship and it is particularly useful to describe the similarity for user pairs with low or even zero meeting frequency. Furthermore, we introduce semantic information on meeting events by using point-of-interest (POI) categorical information. Additionally, when part of the social graph is available as friendship ground truth, we can easily encode such online social network information through a joint graph embedding. Experiments on two real-world datasets demonstrate the effectiveness of our proposed method.