VISUAL SUMMARIZATION AND ANALYSIS OF SPATIO-TEMPORAL DATA

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "VISUAL SUMMARIZATION AND ANALYSIS OF SPATIO-TEMPORAL DATA"

By

Mr. Jiansu PU


Abstract

Increasing amounts of spatial-temporal data is becoming available from 
various kinds of sensors, surveys, and many other sources for researchers 
due to the advances in technologies like GPS, RFID, and wireless 
communication devices. Mining or analyzing these kinds of data can shed 
light into some very interesting applications such as extracting 
population's mobility pattern from mobile data, detecting anomalies from 
vehicle GPS data, monitoring traffic and quickly responding to events by 
mining historical trajectory data, and analyzing peoples' spatio-temporal 
behaviors with social and commercial values from their social check-in 
data. The hidden spatial-temporal patterns in these data can convey useful 
knowledge, which contributes to the decision-making and problem solving, 
and have high social and commercial values. However, analysis of a 
tremendous amount of spatial-temporal data is a very challenging task. 
These data are usually noisy, sparse or incomplete, high dimensional, and 
contain both spatial and temporal attributes. Thus a fast and intuitive 
way to understand and compare these characteristics is indispensable. How 
to clearly summary and explore these complex features in data becomes an 
important problem. It is essential to present the data features in its 
original structure to prevent information loss and facilitate the analysis 
and at the same time provide users with some approaches to characterize 
unique patterns, compare different combinations of features, and quickly 
search anomalies.

Visual analytics has emerged as a very active research field and can be of 
great value for multiple dimensions, spatio-temporal attributes, and 
heterogeneous structures and also provide rich interactions, allowing 
users to explore the data and improve mining processes and results with 
their domain knowledge. It turns complex and abstract data such as mobile 
data, GPS data, historical trajectory, and social check-ins into visual 
representations in intuitive ways with rich context over multiple 
dimensions, and then users can exploit interactive computer graphics 
techniques and human visual capabilities to gain insight into the data. It 
is essential to keep human in the analysis loop to exploit the tremendous 
pattern-recognition capability of the human visual system. And the 
analyst's sense of the space and time, the knowledge of space/time's 
inherent properties and relationships, and space/time-related experiences, 
are hard to convey to machines. Targeting on the visual summarization and 
analysis in spatial-temporal patterns in big data, in this thesis, we 
propose a set of visual summarization approaches on multi-dimensional 
spatio-temporal data, which cover different aspects of data visual 
analysis issues.

•	We propose a visual analytic approach, which integrated many 
well-established visualization techniques such as parallel coordinates and 
pixel-based representations to characterize data's mobility-related 
features and summarize user groups inferred from the results.

•	We develop a novel visualization method, Voronoi-diagram-based 
visual design to reveal the unique features related to flow in the data. 
This visualization method can better reveal the direction information when 
comparing two adjacent flows of time-series data in a graph.

•	We propose a new visual aided mining approach, Visual 
Fingerprinting (VF) for extremely large-scale spatio-temporal feature 
extraction and analysis. The approach integrated important statistical and 
historical information and can be conveniently embedded into urban maps. 
The sophisticated design of the visualization can better reveal frequent 
or periodic patterns for temporal attributes.

•	We develop an interactive visual analytics system, T-Watcher, for 
monitoring and analyzing complex traffic situations in big cities via taxi 
trajectory data. Several new integrated traffic fingerprinting designs 
have been elaborated. We also designed a novel visual structure called 
cell-glyph to compare instantaneous situations with statistical 
information.  The system consists of three major modules (the region 
fingerprint, the road fingerprint, and the vehicle fingerprint) and users 
are able to utilize the carefully designed visual structures to monitor 
and inspect data interactively from three levels (region, road and vehicle 
views) for traffic patterns analysis and abnormal behaviors detection.

•	We present a visual analytics system, Social Check-in 
Fingerprinting (Sci-Fin), to facilitate the analysis and visualization of 
social check-in data. We focus on three major components of the check-in 
data: location, activity, and user. Visual fingerprints for region, 
activity, and user are designed to intuitively represent high-dimensional, 
spatio-temporal attributes related to these components. Some well 
established visualizations like WorldMapper and Voronoi Treemap are 
integrated into our glyph-like designs. The visual fingerprint designs 
allow easy comparison of different check-in locations, different 
activities and user groups, which means they can be conveniently overlaid 
into maps and embedded into graphs and charts.

•	Finally, we demonstrate the effectiveness and usability of our 
methods by conducting case studies on real datasets including mobile phone 
data, taxi GPS data, and social check-in data. Some interesting findings 
have been obtained.


Date:			Tuesday, 6 August 2013

Time:			3:00pm – 5:00pm

Venue:			Room 3501
 			Lifts 25/26

Chairman:		Prof. Peter Chen (ACCT)

Committee Members:	Prof. Lionel Ni (Supervisor)
 			Prof. Huamin Qu (Supervisor)
 			Prof. Lei Chen
 			Prof. Long Quan
 			Prof. Furong Gao (CBME)
 			Prof. Weijia Jia (Comp. Sci., CityU)


**** ALL are Welcome ****