The proliferation of publicly accessible urban data provide new insights on various urban tasks. A frequently used approach is to treat each region as a data sample and build a model over all the regions to observe the correlations between urban features (e.g., demographics) and the target variable (e.g., crime count). To define regions, most existing studies use fixed grids or pre-defined administrative boundaries (e.g., census tracts or community areas). In reality, however, definitions of regions should be different depending on tasks (e.g., regional crime count prediction vs. real estate prices estimation). In this paper, we propose a new problem of task-specific city region partitioning, aiming to find the best partition in a city w.r.t. a given task. We prove this is an NP-hard search problem with no trivial solution. To learn the partition, we first study two variants of Markov Chain Monte Carlo (MCMC). We further propose a reinforcement learning scheme for effective sampling the search space. We conduct experiments on two real datasets in Chicago (i.e., crime count and real estate price) to demonstrate the effectiveness of our proposed method.

Taxi demand prediction is an important building block to enabling intelligent transportation systems in a smart city. An accurate prediction model can help the city pre-allocate resources to meet travel demand and to reduce empty taxis on streets which waste energy and worsen the traffic congestion. With the increasing popularity of taxi requesting services such as Uber and Didi Chuxing (in China), we are able to collect large-scale taxi demand data continuously. How to utilize such big data to improve the demand prediction is an interesting and critical real-world problem. Traditional demand prediction methods mostly rely on time series forecasting techniques, which fail to model the complex non-linear spatial and temporal relations. Recent advances in deep learning have shown superior performance on traditionally challenging tasks such as image classification by learning the complex features and correlations from large-scale data. This breakthrough has inspired researchers to explore deep learning techniques on traffic prediction problems. However, existing methods on traffic prediction have only considered spatial relation (e.g., using CNN) or temporal relation (e.g., using LSTM) independently. We propose a Deep Multi-View Spatial-Temporal Network (DMVST-Net) framework to model both spatial and temporal relations. Specifically, our proposed model consists of three views: temporal view (modeling correlations between future demand values with near time points via LSTM), spatial view (modeling local spatial correlation via local CNN), and semantic view (modeling correlations among regions sharing similar temporal patterns). Experiments on large-scale real taxi demand data demonstrate effectiveness of our approach over state-of-the-art methods.

A wide range of modern web applications are only possible because of the composable nature of the web services they are built upon. It is, therefore, often critical to ensure proper functioning of these web services. As often, the server-side of web services is not directly accessible, several log message based analysis have been developed to monitor the status of web services. Existing techniques focus on using clusters of messages (log patterns) to detect important system events. We argue that meaningful system events are often representable by groups of cohesive log messages and the relationships among these groups. We propose a novel method to mine structural events as directed workflow graphs (where nodes represent log patterns, and edges represent relations among patterns). The structural events are inclusive and correspond to interpretable episodes in the system. The problem is non-trivial due to the nature of log data: (i) Individual log messages contain limited information, and (ii) Log messages in a large scale web system are often interleaved even though the log messages from individual components are ordered. As a result, the patterns and relationships mined directly from the messages and their ordering can be erroneous and unreliable in practice. Our solution is based on the observation that meaningful log patterns and relations often form workflow structures that are connected. Our method directly models the overall quality of structural events. Through both qualitative and quantitative experiments on real world datasets, we demonstrate the effectiveness and the expressiveness of our event detection method.

Recent advances in positioning technology have generated massive volume of human mobility data. At the same time, large amount of spatial context data are available and provide us with rich context information. Combining the mobility data with surrounding spatial context enables us to understand the semantics of the mobility records, e.g., what is a user doing at a location, e.g., dining at a restaurant or attending a football game). In this paper, we aim to answer this question by annotating the mobility records with surrounding venues that were actually visited by the user. The problem is non-trivial due to high ambiguity of surrounding contexts. Unlike existing methods that annotate each location record independently, we propose to use all historical mobility records to capture user preferences, which results in more accurate annotations. Our method does not assume the availability to any training data on user preference because of the difficulties to obtain such data in the real-world setting. Instead, we design a Markov random field model to find the best annotations that maximize the consistency of annotated venues. Through extensive experiments on real datasets, we demonstrate that our method significantly outperforms the baseline methods.

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?

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.

The development in positioning technology has enabled us to collect a huge amount of movement data from moving objects, such as human, animals, and vehicles. The data embed rich information about the relationships among moving objects and have applications in many fields, e.g. , in ecological study and human behavioral study. Previously, we have proposed a system MoveMine that integrates several start-of-art movement mining methods. However, it does not include recent methods on relationship pattern mining. Thus, we propose to extend MoveMine to MoveMine 2.0 by adding substantial new methods in mining dynamic relationship patterns. Newly added methods focus on two types of pairwise relationship patterns: (i) attraction/avoidance relationship, and (ii) following pattern. A user-friendly interface is designed to support interactive exploration of the result and provides flexibility in tuning parameters. MoveMine 2.0 is tested on multiple types of real datasets to ensure its practical use. Our system provides useful tools for domain experts to gain insights on real dataset. Meanwhile, it will promote further research in relationship mining from moving objects.

Movement data have been widely collected from GPS and sensors, allowing us to analyze how moving objects interact in terms of space and time and to learn about the relationships that exist among the objects. In this paper, we investigate an interesting relationship that has not been adequately studied so far: the following relationship. Intuitively, a follower has similar trajectories as its leader but always arrives at a location with some time lag. The challenges in mining the following relationship are: (1) the following time lag is usually unknown and varying, (2) the trajectories of the follower and leader are not identical, and (3) the relationship is subtle and only occurs in a short period of time. In this paper, we propose a simple but practical method that addresses all these challenges. It requires only two intuitive parameters and is able to mine following time intervals between two trajectories in linear time. We conduct comprehensive experiments on both synthetic and real datasets to demonstrate the effectiveness of our method.

With the development of positioning technology, movement data has become widely available nowadays. An important task in movement data analysis is to mine the relationships among moving objects based on their spatiotemporal interactions. Among all relationship types, attraction and avoidance are arguably the most natural ones. However, rather surprisingly, there is no existing method that addresses the problem of mining significant attraction and avoidance relationships in a well-defined and unified framework. In this paper, we propose a novel method to measure the significance value of relationship between any two objects by examining the background model of their movements via permutation test. Since permutation test is computationally expensive, two effective pruning strategies are developed to reduce the computation time. Furthermore, we show how the proposed method can be extended to efficiently answer the classic threshold query: given an object, retrieve all the objects in the database that have relationships, whose significance values are above certain threshold, with the query object. Empirical studies on both synthetic data and real movement data demonstrate the effectiveness and efficiency of our method.