Eurographics Conference on Visualization (EuroVis) 2016

K-L. Ma, G. Santucci, and J. J. van Wijk

(Guest Editors)

Volume 35 (2016), Number 3

RoutineMap: Patterns of Life in Spatiotemporal Visualization

C
Figure 1: RoutineMap integrates and presents the spatial, temporal, and thematic information in a single view. It visualizes peoples daily routines
on the map through utilizing their vehicles GPS data. The concentric dodecagons can represent activities in a single location for (A) a group of
subjects over a single period of time, or (B) a single subject over multiple periods. On top of that, related events such as credit card transactions are
shown, representing amounts by different dot sizes. If a location is labeled, a corresponding symbol is put in the center of the glyph. A: an overview
of the one-subject-multiple-period mode. This view displays elementary information for one persons daily routine with clear office hours, favorite
restaurants, credit transactions, and driving paths. B: a portion of one-subject-multiple-period view for one subject. Other than regular office hours,
this person also loves to go to a bar nearby at around 8pm. C: a portion of one-period-multiple-subject mode. Everybody comes to office between
8am to 12pm, left for lunch, then back from 2pm to 5pm. D: A bald eagles movements in June 2014.
Abstract
Spatiotemporal data are common in visual analytics tasks, but identifying events over time and geographical locations in a
single view is commonly known to be challenging. Previous work has addressed part of the problem, but it is still difficult to
identify the routines or patterns through the visualized space, time, and attribute values immediately and intuitively. This
paper introduces a glyph-based visualization approach to integrate time and space information concurrently and focus on
activities at the comparatively sparse fixed points locations. Based on the analysis of different design options, we have
developed a set of polygon-based glyphs to depict time, and utilized the center of polygons to show the location. With thick
orange segments to indicate stays and color dots to represent activities on the polygon, this approach is appropriate to
explore temporal events in the spatial context. We demonstrate how this technique can be applied to the GPS data and
calendar event data. The evaluation obtained from VA experts and analysts indicates that the polygon-based glyph
visualization is able to detect temporal and spatial patterns and conclude life routines in a clear and intuitive way.
Categories and Subject Descriptors (according to ACM CCS): H.5.2 [Computer Graphics]: User InterfacesGraphical User
Interfaces (GUI)

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

1. Introduction
Nowadays, the embedded GPS in mobile phones or tracking
devices can produce detailed spatiotemporal data including
locations, time stamps, trajectories, and may even record events
for the moving subjects. Visualizing such spatiotemporal data is
always crucial for analysts to identify events and behaviour
patterns over time. According to Peuquet [Peu94], three key
components in spatiotemporal data are space (where), time
(when), and subjects (what). Where indicates the spatial
reference information, when is the time, and what refers to
the subject. Their inherent properties and intrinsic relationships
are full with variations and contain rich information. Through
aggregating a series of unfolding events, visualizations are able
to help analysts to interpret event sequences, define the history,
and derive patterns and knowledge from the data. Instead of
visually analyzing where, what, and when separately, it is
essential to display spatial and temporal components in an
integrated view [SMZZ13]. Therefore we developed
RoutineMap, a novel visualization approach using polygon
glyphs to integrate and show spatial, temporal, and event
thematic data in a single view. The intuitive and unique ability
of the novel approach facilitates exploration and identification
of spatiotemporal patterns across both the spatial and temporal
dimensions. The main idea of the paper is to use a glyph-based
approach to provide fluent and flexible use of visualization for
places (where), stays (when) and subjects (what)
[HS12]. The polygon-based glyphs intuitively map the
multivariate attributes including a given subject, time periods of
stay, the parking location, and what happened during the time
that the subject occupied the location on a map. Several
polygons can be nested together on the map to present a set of
subjects, a set of stays, the parking locations, and events
occurring at times while the subjects occupy places.
RoutineMap is suited for comparatively sparse fixed pointsbased locations where the interest is focused on activities at
those points. The scalability leads to a generalization of the
glyph-based approach to work with many thousands of
locations.
RoutineMap was built upon the previous work of [Guo14],
which won an award in a major international contest in Visual
Analytics. After the competition, we conducted analyst
evaluations, revised and enriched the work significantly. It
warrants this paper to thoroughly discuss the new visualization
design and contribute in the following aspects:

Provide a unique glyph-based approach to integrate and

visualize when, where, who, and what information
concurrently in a single view for spatiotemporal data.

Introduce the use of 12-sided polygon rather than circle

to aid precise reading of the clock time.

Directly reveal peoples activity patterns and daily

routines through nested adjacent days or subjects.

With simple interactions, the analyst is able to

investigate patterns and drill down for details in event-related
changes.

With simple change, the solution can be scaled up to

accommodate much higher numbers of subjects, or extend to
different time periods.

Generalization of the design to visualize habitats for bald

eagles.

information from both spatial and temporal dimensions.

Minimally, a GPS tracking dataset has three dimensions a
timestamp with Latitude and longitude to record the
position at time on a planar map. However, due to the
limitation of the 2D screen, many visualization techniques
must focus on one single dimension, either time or space,
and visualize the other dimensions as supplements. For
example, many researchers analyze time series data with
curves [PMRS96][Kar94][KKA95], stacked area charts
[HDKS05][HHN00], and small multiples [TG83] to show
the value change over time. But the information of spatial
locations has to be simplified as labels that visually do not
provide any spatial meaning. Such solutions challenge
analysts cognitive capabilities to mentally connect, bridge
and interpret information from many views. To make the
problem more complicated, the GPS data can often be
enriched to contain thematic (nonspatial) attributes of a
subject, such as speed, or events like credit card
transactions and phone calls. It is difficult to analyze and
compare time with a spatial dimension [AAG03]. With the
added multiple variables in the visualization, it will be far
more complicated for analysts to detect routine changes.
How to help the analysts to discover events and detect
undergoing changes of spatial properties, even thematic
properties? Animation is a typical way to detect differences
in time. Interactive multiple views is another strategy to
display the state of information of each moment and to
support multivariate analysis. However, because of a
humans limited short-term memory and change blindness,
understanding these representations may become a
laborious cognitive process and impose high cognitive
overload on viewers.
It is obvious to use a map to denote spatial variables.
But how can we integrate the temporal dimension on the
map? Researchers have developed various methods to
integrate temporal information, mainly including 2D and
3D timelines [KW04], animated timelines [PMRW*96],
time slider bars and labels [SH10][PVF13] over the map.
3D solutions such as GeoTime [KW04] work extremely
well to show a small amount of moving objects, but will
suffer severe problems of visual cluttering when the data is
getting bigger. Scheepens et al.s study [SWWW11]
creates a density map of vessel traffic to show an aggregate
overview of large amounts of data. Nevertheless, without
showing the history before the selected time, it challenges
the analysts visual perception to understand the complex
visual space. Spatial and temporal data are equally
important and closely linked in terms of activity pattern
analysis, so it is critical to visually present both of them in
one view for more efficient and effective interpretation.

To maximally augment analysts cognition and reduce

cognitive load, a visual analytics system should be easy to
read, easy to understand, and easy to remember [HEH09].
Adding a timeline dragging bar is easy and common, but
how to record and provide an effective history overview
without interrupting the other visualizations is difficult. A
famous infographic example is Napoleons Russia
campaign in 1812, designed by Charles Joseph Minard
[TG83]. It is successful and effective because of the
combination of different dimensions: location, time,
temperature, loss of life, and historical context. Inspired by
this combing and recording approach, we developed
2. Motivation
this novel solution RoutineMap to analyze spatiotemporal
In our life, many events happen at only certain places at certain
traffic problems and to visualize the movement history on a
time. To fully understand the data to detect patterns and map.
anomalies, it is crucial to let the analyst see and connect

combinations of visualizations and coordination. Some

researchers have focused on dynamic representations,
3. Related Work
mostly animation, to display changes over time
3.1. Temporal data visualization
[RFFL*08][APP11][FQ11]. It is a natural approach to
representing temporal information and detecting subject
Time is an outstanding dimension according to Shneidermans
data-type taxonomy [Shn96]. Researchers have developed movement patterns.
various visualization techniques to map different types of time.
Although these different techniques have shown their
Classic approaches represent time as a horizontal axis and map ability to synchronize the presentation of multivariate
data values to the y axis using line, scatter, stacked bar, bubble, spatiotemporal data, analysts may suffer from a heavy
and theme river [TA00]. Although these techniques are good to cognitive load since they have to mentally remember and
show the overall patterns of the data, they are not ideal to connect information from different views, which makes
facilitate comparison of values. To solve the overlapping difficult to detect changes in the visual representation and
problems in the collections of time series data, Tufte [TGm83] efficiently compare different attributes.
proposed the concept of small multiples to compare different
small representations of a data set. However, it is still very
3.3. Temporal event sequence visualization
difficult for analysts to summarize events and trends when the
Events are important when analyzing humans life pattern.
number of time series grows large [Kei00].
An event happens at a given time and location, lasts a
In order to show periodical temporal patterns, a common
certain period of time, and contains several subjects.
approach is to use circular layout to map temporal information
Timeline is a natural way to display time events and
on co-centric rings or spirals to show daily, weekly, and yearly
provide an overview of event sequences. Visualizing
patterns. Keim et al. [KSS04] divided a circle into several
multiple records in parallel has become a popular technique
segments and subdivided each segment into time slots to
for presenting trends in the data [WG12]. Lifelines
visualize fast changing time-dependent values. Carlis and
[PMRW*96] use horizontal lines and icons to represent
Konstan [CK98] found that many systems used timeline as a
patient records and youth criminal histories. Time Manager
basis to display time attributes; however, none of these supports
[AAG00] connects to a dynamic map and allows analysts
the exploration of the periodic aspects of the data. They also
to select time moments or internals to support different
stated that spiral visualization is easier to detect patterns in the
analysis tasks. Systems like LifeFlow [WGPW11] and
serial periodic data compared with traditional linear displays.
OutFlow [WG12] aggregate multiple event sequences and
SpiraClock [DH02] is another similar example to map temporal
combine them into a tree. Hagerstrand [Hg70] is the first
data in a spiral layout. Analysts can anticipate nearby events
one who proposed the space-time cube concept. His
with the white spiral inside an analogy clock. Fischer, Fuchs &
framework has enlightened researchers to encode time as
Mansmann [FFM12] introduced a novel combination of 24-hour
the third dimension and explore event sequences with a 3D
clock-based glyphs with a circular tree map to visualize
timeline. Kapler and Wright [KW04] placed spatial time
hierarchical time-series data. They also reported that the star
tracks and 3D z-axes timelines to perceive when and where
glyph and the clock glyph are more effective to detect temporal
on a 2D map. But the 3D view may not scale well for large
locations rather than the line glyph [FFMB*13]. All these
amount of subjects. Researchers often use techniques such
approaches show that radial mapping of temporal information is
as filtering, grouping, and clustering to explore event
able to support periodic structures in the data. Radial glyphs
sequences. Filtering and querying are suitable for looking
make it easy to compare multiple geospatial attributes as well as
up subjects and attributes. Grouping and clustering aim to
scale to large data sets.
find close subjects or groups and classify events based on
3.2. Integrated views of spatiotemporal mapping
their similar patterns. Alsallakh et al. [Als14] segmented
the respective time series data into intervals that
Spatiotemporal mapping has long been a challenging research
corresponded to the constituent activities and compared
problem in information visualization. How to answer the three
them for analysis. Trajectories of moving subjects are
questions what, when, and where concerning spatiotemporal
observed as a chronological order of stops and moves
data has been widely studied. Andrienko et al. [AAP03]
[Xiu11]. Comparing routes and dynamics is more complex.
discussed three basic cartographic depiction models single
Various approaches, such as Pelekis et al. [PKMN*07] and
static map, multiple static maps, and animation map. The widely
Rinzivillo et al. [RFFL08] have suggested different
known standard representation for single static map is flow lines
solutions.
or arrows drawn on a map [Vas97]. Minards map of
Napoleons march to Moscow [TGm83] is a classic example of
A major challenge of spatio-temporal visualization is to
using storylines on the map. Similar approaches can be found in develop representation and strategies that combine spatial,
[SMZZ13][AAW07][TM12]. None, however, integrate accurate temporal and other attributes simultaneously and closely to
geographic information with time and subjects, making it help analysts to identify patterns and event sequences
difficult to read and analyze. Guo et al. [GCML06] drew effectively and efficiently. In RoutineMap, we combine
attention to detecting geographic, temporal, and multivariate time, subjects, and events all together as dodecagon glyphs
patterns in multiple ways, including a self-organizing map, a on a map. Within a single simple visualization, analysts are
parallel coordinate plot, several forms of reorderable matrices, a able to explore, interpret, synthesize, and analyze the
geographic small multiple display, and a 2-dimensional geospatial data in an intuitive graphic interface.
cartographic
color
design
method.
Some
authors
[WDSC07][SDWC07] have applied the mash-up technique to
4. RoutineMap
interactively aggregate data by space, time, and attributes.
Fredrikson et al. [FNPS99] created aggregates together with the
4.1. The GPS tracking data
SnapTogether visualization system to display large quantities of
A GPS tracking dataset typically contains: timestamp,
data. Through temporal, geographical, and categorical
latitude, and longitude. It usually records every 1 second
aggregations
of data points, analysts can implement different

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

even when the subject stops. Some GPS trackers are smart
enough to trigger data recording only when the subject is
moving. A simple GPS tracker may record data periodically
until it runs out of power no matter the subject is moving or not.
If the GPS tracker does not have new records for a prolonged
time (e.g. > 60 seconds), or the location does not change for a
while, it implies that the subject currently stops at this location.
It is easy to identify each GPS with a unique id in terms of
different GPS tracking devices. Thus a basic data set with
several GPS devices may look like:

data analysis and information needs of analysts. In GPS

tracking data, a basic question could be where are the
subjects at different time. Ideally, it would be great if we
can answer the question in terms of multiple subjects
during multiple periods. To make the solution feasible, we
distinguish the big question into two sub-tasks:

Patterns for one subject in several periods (Fig. 1a)

Patterns of many subjects in one period (Fig. 2).

Though for each mode the data components are

different, they all integrate the three dimensions of data
space (where), time (when), and subjects (what). Bertin
Table 1: a GPS tracking data extracted from a commercial GPS [Ber10] also distinguishes three levels of reading in the two
device
basic questions: elementary level (with respect to a single
ID
Time
lat
Lon
item), intermediate (with respect to a group of elements),
1
2015-06-27T23:04:06Z
46.2590
and overall (with respect to all items together). An analyst
1
119.266
may also need to compare sets of subject movement history
1
2015-06-27T23:04:07Z
46.2587
during one day or examine a single subjects movement
history during a certain period. For life pattern analysis, we
9
119.266
identify the following types of questions in two modes:
We can further enrich the GPS tracking data by associating it
Questions in one-period-multiple-subject mode:
with events. Many events can be tracked by their time. If an

When and where is a particular subject presented

event is associated with a moving subject, we can easily find out
around the location? (Elementary level)
where the event occurs through matching the time.

What are the trajectories of the subjects in the

Many researchers have studied human mobility patterns
period? (Elementary level)
using different methods. Do and Gatica-Perez [DGp14]

What subjects are presented at the selected period?

collected smartphone data over 18-months in late 2009 in
(Intermediate
level)
French-speaking Switzerland. The data reasonably reflect actual

What subjects and in which places are presented at

analyst mobility trends and show that a large fraction of people
visit from 2 to 4 places per day. Another study [YKBN14] also a particular moment/during an interval? (Intermediate level)

What are the attributes related to the event (what

confirms that approximately 90% of people visit less than 5
places per day and the most regular visited places are likely to happens for the subjects at the location during the given
be home and workplace. According to Bertin [NSMP11], time)? (Intermediate level)

What are the similarities and differences between

research has been conducted to analyze the dynamics of
Foursquare analyst activities and the findings suggest that there two subjects during the period? (Intermediate level)

What is the overall behaviour of a group of subjects

exist two different patterns during weekdays and weekends.
People usually go to work in the morning, eat lunch and return during the period? (Overall level)

Questions in one-subject-multiple-period mode:

home between 6pm to 8pm or go to malls and bars in the
evening. At weekends people tend to have more leisure related

What are the similarities and differences between

activities in a restaurant or a hotel between 12pm and 10pm. It is two periods for the subject? (Intermediate level)
obvious that human mobility patterns are regular and mostly

What are the routine patterns of the subject during

predictable in the real life.
all the periods (Overall level)

What is the trend of the subject over the days?

The simulation data from the VAST 2014 Mini-challenge 2
has a high fidelity on spatial and temporal dimensions. It (Overall level)

What is the trend of the subject across the

captures the key characteristics of humans life patterns and can
be representative of the real daily activities. The characteristics locations? (Overall level)

How did the subject movement and parking

of the data match the activity patterns from many empirical
patterns
change
over the entire time period? (Overall level)
studies of human mobility. We also conducted a small scale
study by tracking several people for a short period of time. The
characteristics of the tracking result are very similar to the
VAST data. In the following discussion, we will use the VAST
data to present and discuss RoutineMap since it is a
representative enough data set and avoids invading subjects
privacy.
The GPS tracking data in the Vast Challenge dataset contains
40 cars and trucks with GPS trackers for a two-week period. Car
drivers drive their own cars, while two or more drivers share one
truck. Each GPS has its unique ID. The data have 70 million
GPS records with geographic coordinates, timestamps and the
vehicle IDs, as well as 1500 credit card transaction and 1400
loyalty card transaction data including records with timestamps, Figure 2: One-period-multiple-subject mode to show the routines of
store names, price, and the drivers names.
all subjects in one day
4.2. Data analysis and information needs

Depending on the nature of human activities in the

modern
life, the periodical pattern may show up in terms of
From the data structure point of view, Bertin [Ber10] identifies
two basic question types and three distinct levels to describe hour, day, week, month, and year. Here we choose to study

the daily patterns since most human activities are repeated on a car(n) (Fig. 4a) or day 1 to day (n) (Fig. 4b). The red color
daily basis. Through the daily patterns, we can easily identify dots on the rings show who and when carry out the activity
the weekly patterns.
at this location. The polylines connecting the centers of
dodecagons show the subjects trajectories. The nested
To gain a thorough understanding, the analyst can switch
dodecagons define certain relationships of groupings or
from one mode to another, compare a set of subjects on one day
occurrences. They make it easier to perceive changes and
or identify daily patterns of a single subject. The one-periodidentify event sequences over time in the spatial context. In
multiple-subject mode answers elementary level questions while
Fig. 4(a), eight subjects went to a restaurant during the day.
the one-subject-multiple-period mode solves the intermediate
Three had lunch at around 12:30 pm, six were there from
and overall analytics problems. They supplement each other to
6:30 pm to 10:00 pm after their work. Among these people,
describe a much larger scope of the three attributes when,
two subjects went there twice for lunch and again for
where and what.
dinner. The red dots show that they paid the bill by credit
card at the end. Fig. 4(b) tells us that the person starts to
4.3. A dodecagon to integrate when, what, and where
work at around 8 am, leaves at about 12:00 pm, and then
comes back to work from 2 pm to 5:30 pm during
weekdays. There is a gap between the polygons showing
that she/he never works on weekends. We can also notice
that she/he had a credit card transaction at 12:00 pm on
Monday in the office.

Figure 3: One dodecagon is able to show space (where), time (when), and
event (what). We can tell that the subject goes back home twice from 5:59
pm to 6:50 pm and 8:40 pm to 7:50 am.
Inspired by the clock metaphor, we choose to use a closed nsided regular polygon to represent one period. With marks on its
sides and locating the polygon on the map, it can naturally
integrate information of time, space, and subject, making it
possible to identify the patterns and events at the first glance
(Fig. 3). To represent one day, we map 24 hours to the
dodecagon: each side represents two hours. The vertices can
help us to distinguish different hours. The thick orange segment
on the outline of the dodecagon denotes the time range of stay at
the current location with start time and end time. When the
subject stays at one location without moving for a period (we
choose 1 minute in this dataset), we consider this location as a
Point of Interest (POI) and place one dodecagon at the location
with the time of stay marked with orange segments. The person
may stay at the same location for at different time periods,
which will result in multiple segments on one dodecagon. The
color dots show activities that happened at this location such as
Twitter messages or credit card transactions. The dots radius
can be used to encode values such as the number of tweets
created or the amount of money spent on the credit card. Using
the dodecagon glyph, the analyst can easily see when, where,
and what happened. The dodecagon has a logo in the middle
where the location types are shown. Different logos correspond
to the different location types (e.g., a caf shop, a bar, an office
building, or a house). By doing so, one dodecagon is able to
display the two attributes: time (when), and subjects (what). By
placing the dodecagon on a map, the third attribute where is
presented.
When different subjects stay at the same location in the oneperiod-multiple-subject-mode, or one subject stays at the same
location during multiple days in the one-subject-multiple-period
mode, we use dodecagons with different radiuses to distinguish
different subjects in the day (one-period-all-subject) or different
days for the subject (one-subject-all-period). The radius
corresponds to the subjects ID or a day. When multiple subjects
stay at the same location during one period, or the same subject
stays at the location during multiple periods, these dodecagons
will naturally be nested together to show patterns (Fig.4). From
the inside to the outside, the dodecagon can represent car1 to

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

(a)
(b)
Figure 4: Dodecagons are nested at POIs to show patterns and
events. (a) In the one-period-multiple-subject mode, several subjects
went to a restaurant regularly and paid the bill by credit. (b) In the onesubject-multiple-period mode, the person works in a regular schedule
during weekdays.
4.4. Merging close-by parking locations

Figure 5: Left: before merging. Dodecagons are not co-centric since
the parking location will not be exactly the same for different people.
Right: Nearby locations are put in co-centric.
Extracting POIs is crucial to spatiotemporal data analysis.
If a subject spends a long time in one place, we consider it
significant. A group of subjects may have multiple stays
within a place, or the subjects may periodically visit the
same locations. The repeated stays/stops at POIs can show
a persons daily routine, or discover a relationship among
people since they stayed in one location often. They are
key elements for analyzing and comparing entities
behaviors. A dodecagon is drawn on the map if a subject
stops at one place for more than a certain period. But
she/he will not usually stay at the exact same place. If we
use the original location values, the dodecagons will be
placed like Fig. 5 left. We can see that many stops are very
close. These rings are nested and close to concentric. If the
two locations are very close (e.g. different spots in the
same parking lot), in reality, they should be considered as
one place. Although we can still see the temporal patterns

of stays and events, we merge the close-by locations (50 meters

in this data) into one (Fig. 5 right) in order to make the patterns
on the rings easy to read. The center of the new location is a
simple average of all locations within 50 meters. Thus the
nested rings become co-centric and the patterns of stay and
activities become more clear and obvious.
4.5. Interaction
The main interface of RoutineMap is Fig. 6. By default it shows
multiple cars in one day. The analyst can easily switch back and
forth from the multiple-car-one-period mode to the one-carmultiple-period mode by clicking on the calendar icon on the
left panel (Fig. 1a). RoutineMap not only provides an effective
overview and freedom of zooming in or out, but also supports
various interactions that allow analysts to filter and examine
attributes in detail.

Figure 6: The main interface. Default shows multiple cars in one day.
In the multiple-car-one-period mode, it is very easy to notice
the common patterns of cars through seeing that [MG90]. We
can answer elementary-level questions such as when the
employees leave their houses and return, where they have
breakfast, lunch, and dinner, and which driver works overnight
hours. Furthermore, it is easy to capture significant social events
such as when all the executives have a party (Fig. 7a) and who
are socially very close from different departments (Fig. 7b).
While mousing over a dodecagon, it will be highlighted, and
other dodecagons that are related to this one (the same car in
multiple-car-one-period mode, or the same day in one-carmultiple-period mode) will also be highlighted (Fig. 1 a). It
provides the convenience for the analyst to capture the detailed
route for that particular car or day. By doing so, it is clear to see
the incoming and outcoming trajectories of a dodecagon and
examine different visits belonging to the same subject or the
same day. Since trajectories do not reflect their directions,
analysts can understand the ordering through reading the
temporal information on the highlighted rings. For example,
based on the orange segments on highlighted rings, analysts can
tell that a person visited a coffee shop at 8 am, went to work at
8:30 am, stopped at a restaurant at 12:30 pm, went back to work
at 1:30 pm, visited the same restaurant at 7 pm and went home
at 10 pm. Thus, it is very easy to see related daily activities and
differentiate if people went for dinner directly from work or
after visiting their homes.
The analyst can compare two different subjects in one period
(one-period-multiple-subject mode) or compare two days for the
subject (one-subject-multiple-period mode). To do so, the
analyst can click on a dodecagon to permanently highlight
related dodecagons (denoting the same subject or the same day).
Then the analyst can move over another dodecagon to use a
different color to highlight another set of dodecagons for a
different subject or a different day, which will allow the analyst
to compare two sets of dodecagons and the trajectories with
different colors.

In one-subject-multiple-period mode, the analyst can

easily identify an individuals daily patterns. For example,
what places does the driver usually visit after work? If so,
how often does it occur? When did the event happen at the
given location? Furthermore, we can see some anomalies,
for example, one person visits another persons house from
1 am to 6 am (Fig. 7c) and one person goes to a car repair
shop every day (Fig. 7d). Therefore a single map view
enables analysts to notice abnormal events occurring in the
spatial context and explore them in detail.

(a)

(b)

(c)

(d)

Figure 7: (a) Five people meet at one location from 12:30 pm to

6:30 pm. Everyone spent some amount of money in the club. (b) Two
subjects lived together, went to a caf shop for breakfast every
morning, and went to another caf shop for dinner several times
during the weekdays. (c) One person visited other peoples home at
3:30 am and then left at about 8 am in the morning twice during two
weeks. (d) Another person visited a car repair shop twice during the
two weeks. He stayed in the shop from 7:40 pm to 8:40 pm and spent
money on Friday.
4.5.1 Zoom in to solve issue of overlapping

Figure 8: Desynchronized zoom in to solve issue of

overlapping among nested rings
In the reality, locations may not distribute evenly on the
map. Some locations may be very close. For example, at
the right corner in Fig. 2 (also zoomed in as Fig. 8 left),
there will be an overlapping problem across different
groups of dodecagons. The nearby dodecagons may
intertwine with the neighbours and can cause legible
problem. To solve the problem, we use desynchronized
scale rates between the overall view and dodecagons when
the analyst zooms in. We set the scale rate of rings at half
of the one for the map. For example, if the map zooms to a
scale of 2X (twice the previous size), the rings only zoom
to a scale of 1.2X size. Thus, the distances between two
rings are doubled, while the rings themselves only expand

20%, which will make the overlapping rings separated. The has 60 seconds. One line in a dodecagon will correspond to
more the analyst zooms in, the more overlapping will be 5 minutes in an hour or 5 seconds in a minute. Generally
avoided (Fig. 8 right).
speaking, a polygon is suitable to represent different time
units, such as a heptagon for seven days of a week.
4.6. Implementation
6. Evaluation
To test our design, we implement the RoutineMap as a Web
application built on a standard LAMP (Linux, Apache, MySQL,
6.1. Peer review
and PHP) environment. It is implemented in modern Web
Peer review, especially the review coming from experts
browsers using the HTML5, CSS, SVG, JavaScript, jQuery, and
and practitioners, is an important and constructive type of
the D3 visualization toolkit. VAST 2014 MC2 data contain two
evaluation to obtain critique of work. RoutineMap was
weeks of vehicle GPS tracking data and credit and loyalty card
lucky to get expert peer reviewed from domain experts who
transaction data. It requires analysts to identify the
are all PhDs in information visualization and visual
commonalities and unusual events inherent in this data to
analytics. The peer reviewers provided scores and critical
understand the employees' daily life.
reviews using the reviewer guide. The reviewers used
expert judgement to evaluate RoutineMap and provided us
5. Evolvement of the Design
professional comments, and valuable insights into the
The polygon-based glyph design came from an iterative design novel glyph-based visualization approach.
process. We started with a basic idea of representing subjects
stops with circles. If a subject stopped at a certain location for
more than 60 seconds, a circle will be drawn at the
corresponding geographic location (Fig. 9a). The bigger the
circle, the longer the subject stays. Nevertheless, too many
circles are needed to denote the stay. Guided by Tuftes
principle utilizing narratives of space and time [TGm83], one
intuitive approach is to use a clock metaphor to represent
temporal information. Then we use the segments of a circle to
display the stays. Different colors of the circles denote different
time periods in the morning, afternoon, or evening (Fig. 9b).
However, problems came up when we mapped the periods of
stay to the circle. Analyzing and identifying the changes over
time is very difficult. The 12-hour clock can show only the time
periods in half a day. With the purpose of maximizing the dataink ratio [TGm83], we transformed a 12-hour to 24-hour clock
to represent daily time dimensions on a daily basis. The center
of the clock maps to the location where the subject stays. Thus
with one circle, we are able to show the stay and movements in
a day.

All the reviewers considered the visualisation tool

represents the data well, allows for easy analysis and
exploration, and the analysis result can be directly draw
from the visualization. One reviewer said, The structure of
the analytical technique was established by the team as a
delineation of the methodology, making the entire structure
understandable. By providing the definitions of the
graphics, what information they were depicting, how that
information was visually organized, and what analytical
data was obtainable via the information, the depiction
became self-explanatory. Another reviewer stated, This
is a concise, well-presented project which successfully
applies various geospatial and temporal visualisation
methods for the analysis of the provided dataset. It is
easy to switch from an overview to the underlying 'raw'
data, which is essential of one wants to draw reliable
conclusions from the visualisation.

Thus we decided to use regular polygons to represent one period

(one day). One day has 24 hours and one year has 12 months. So
the 12-sided polygon is suitable to represent the 24-hour clock
and 12-month year. Dodecagon has 12-line segments. We can
use each segment to represent two hours in a day or one month
in
a year. Similarly, one hour has 60 minutes, and one minute

We plotted the same data using space-time cube.

Different colors are used to differentiate different
trajectories. Figure 10 shows that space-time cube is able to
visualize subjects movements and stays. The slope of the
trajectory indicates the speed of movement. Less steep

There were also comments about the overall design of

the visualization, which is both attractive as well as
informative. One reviewer reported that the handling of
temporal patterns through the concentric dodecagon rings
is very elegant. The informal evaluations also showed that
the
events
concerning
daily
and
social
activities/relationships, meaningful suspicious activities or
nonsuspicious unusual events could be immediately
observable and understandable to analysts. The system
facilitates
exploration
of
various
data
points
simultaneously, and focuses on particular cars (employees)
or locations. In general, the method is able to provide good
(a)
(b)
evidence and interpretations for judging the nature of
Figure 9: Design changed from a circle view to a clock view: (a) a discovered events.
circle to represent the time periods of stay for one subject; (b) a 12-hour
6.2. Comparison between RoutineMap and 3D spaceclock to display the stay.
time visualization
The clock glyph is intuitive to display time, but hard to
recognize the time points or time intervals accurately. We compared RoutineMap with a current existing
Theoretically, a circle is a polygon with an infinite number of technique called space-time cube in visualizing spatiosides, and analysts can judge the time by looking at the position temporal-thematic patterns. Space-time cube [Hg70]
of the data point on the circle or arc lengths between two data /Geotime is an effective approach to display events and
points. We need to provide some visual aids to facilitate time activities within a single combined temporal and geospatial
reading. A regular approach is to mark small ticks along the display. The Z-axis of the cube denotes the temporal
circle. But with regard to the visualization with so many nested information while the X-Y dimensions represent the spatial
circles, the ticks will make the graph messy and hard to read. information.

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

slope corresponds to trajectories of fast vehicles while steep

segments denote relatively slow speeds. Vertical lines occur
when the vehicle stops at a place for some time period. When
two or more subjects meet at one place, the vertical lines meet
into one. The cube provides an overview of the data set both in
space and time and exposes characteristics of movement
patterns. However, there exist several problems for the cube.
First, the analyst is hard to determine the time and location of
one point on the 3D curve, especially at an arbitrary viewing
angle. Also the cube can only effectively display a small number
of trajectories. Showing 40 trajectories simultaneously makes
the plot messy and unreadable. The overlapping vertical lines
make it hard for analysts to view any spatiotemporal patterns of
events even with rotating the cube to explore more.

afternoon and at night. Fig. 11b is a screenshot of the bald

eagles activities from November 3rd to November 10th.
We can see that it is migrating southward. It usually starts
to travel in the afternoon and takes a rest in the evening.
With this real but imperfect dataset, we are able to
demonstrate that RoutineMap is an ideal tool to study the
life habitat of bald eagle including nesting and migratory
patterns.

(a)

Figure 10: Screenshot of Space-time cube
We can tell that with the current number of subjects (10 to
50), apparently RoutineMap outperforms Geotime with less line
crossing, less visual elements, and better legibility. RoutineMap
is able to provide an intuitive and effective means to see patterns

of life, while in Geotime demands the analyst has to rotate the
(b)
view to find the overlapping vertical segments from different
angles. In terms of visualizing thousands of entities and time Figure 11: one bird in multiple days. (a) Juvenile bald eagle
in June. (b) Juvenile bald
records, neither space-time cube nor the current RoutineMap behavior patterns during the fledging period
rd
eagle
migratory
patterns
from
November
3
to
10th.
can scale well. However, we can use aggregation to handle large
numbers of subjects. Section 6.2 further discusses the scalability
7. Discussion
issue of RoutineMap.
7.1. Different time units to scale
6.3. Generalization to different datasets
In the RoutineMap, we used a 12-sided regular polygon to
Here we use our visualization method to examine a set of real represent a day. The 12 sides are carefully chosen to meet
animal tracking data. The GPS tracking data is for one young the 24-hour cycle of a day. For other time units, for
bald eagle generated from the New Jersey Bald Eagle tracking example, we can visualize a week with a seven-sided
project [CKSS*14] and was accessed through Movebank Data regular polygon (one day per side Fig. 12). The daily
Repository (https://www.movebank.org). It includes specific pattern is maybe a little bit vague, but the weekly pattern is
timestamps, latitude, and longitude of a juvenile bald eagle definitely clear. Twelve months of a year can also be
(about 8.5 weeks old) who was fitted with a GPS tracking visualized by a dodecagon. But for monthly events, since
device from 05/22/2014 to 12/03/2015. Unlike our previous the length of each month varies from 28 to 31, using a
simulation data, the animal tracking GPS data are in irregular circle could be better.
log intervals ranging from every a few seconds to minutes.
Fig.11 provides a holistic view of the eagles activities in
June and November in 2014. It is in the one-eagle-multiple-days
mode. A dodecagon is plotted if the bald eagle stays at one place
for more than 5 minutes. Within 0.1 mile, the centers of the
close dodecagons are merged. We can see a clear spatiotemporal
pattern of the eagles daily activities from Fig.11. In June, the
eagle sleeps in this nest during night time (bottom right corner
of Fig.11a). It also loves to stay at one place close to the nest.
Before June 10th, the bald eagle remains around the nest
because it is still young (about 10 weeks). After June 10th, it
starts to take short flights to some nearby places and return to
the nest tree frequently. Some frequent flights were made
southeast in the late morning. It also frequently visited two
places on the north of the nest and spent considerable time in the

Figure 12: 40 subjects weekly routine patterns. A heptagon
represents one week. From inside to outside are different subjects.
7.2. Scalability
When there are a large number of subjects, eventually the
radiuses of the nested dodecagons will become very large,


which will cause severe overlapping problem or legibility
problem. It will be hard to discern groups and patterns in the
mass (Fig. 13 top). We can reduce the space between rings, but
considering humans limited visual acuity and monitor
resolution, we cannot make the dodecagon too small, and we
have to keep the gaps among dodecagons.

Figure 13: Scale to many subjects with aggregation. Top: 80 subjects.
Bottom: Nested rings are compressed into one ring with color shade still
show the frequency of stay clearly.
One way to solve this scalability problem is using
aggregation to fill all subjects in to one dodecagon (Fig. 13
bottom). We can use color shades to encode the percentage of
stays during the time periods. Darker colors such as deep
yellows represent frequent stays during the time, while lighter
colors represent less stays. Thus we can make dodecagons to be
flexible to handle large data sets and show daily spatiotemporal
information. They can preserve the overview of the large data
sets. But this method can lose the detail information for
individual subjects. Analysts can zoom in or click on one thick
ring to expand it as nested rings.
7.3. Other limitations and challenges
RoutineMap emphasizes temporal comparison at the expense of
spatial fidelity, or spatio-temporal details at the expense of
attribute comparison. However, there are still some information
that cannot be covered by RoutineMap, for example, speed. It is
best employed when interval or discrete temporal information is
important, but not ordinal time scale. It is not suitable to
visualize more than 50 subjects because overlapped glyphs may
lead to visual occlusion and ambiguity for massive amounts of
information.
Another limitation of RoutineMap comes from the nested
rings. Rings can be nested to represent adjacent days or subjects
but within a limited scale range. The inner rings are smaller and
therefore have lower precision and lower visual importance.
More space is given to the outer rings than to the inner rings.
Thus, it is easier for people to discern details in larger rings.
Based on our design practice, the smallest radius of the polygon
for detecting the temporal information should be no less than 20
pixels.
New user of RoutineMap has the challenge to read the 24hour time quickly from one polygon/circle. We have been so
used to 12-hour clock from daily life so that it will take some
warm-up time to get use to the new setting.
8.

Conclusion

This paper presents a novel polygon-based glyph approach, a

visualization that enables flexible and efficient analysis of
places (where), stays (when), and subjects (what) in a

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

single map view, to allow analysts to intuitively read the

patterns of life (human or animal). In this work we explore
the different radial layout glyphs to portray the spatial and
temporal data vividly and accurately. We find that a 24hour dodecagon glyph provides a robust and efficient
technique to identify stays and movements in an easy and
clear way. Implementing the glyph with polygons
containing different sides makes it scalable for most of the
time and dates. Also, the glyph-based visualization assists
human recognition of complex spatiotemporal information
in a natural and intuitive way. The nested polygons
aggregate multiple-event sequences and thus help analysts
to gain insights into overwhelming events involving time,
subject, and location concurrently. It further leads to the
discovery of trends and patterns in details. The novel
approach solves the information overload problems and
reduces visual complexity in large quantities of
spatiotemporal data through scaling the glyph up or down
to different levels of data. The effective interaction also
facilitates analysts to explore events and trends in the
datasets. With different interaction and aggregation
strategies, our solution can be scaled to accommodate
much large number of subjects, longer time period, or
different time units. The solution is effective in terms of
not only helping to detect anomalies from simulated
competition datasets, but also identify animals nature life
patterns from real nature datasets. Through extending the
application in different datasets, we are confident to claim
it as a very promising new visualization method.

Acknowledgements
We thank all the participants who participated the analytics
study. Many thanks to all competition reviewers for their
constructive comments and valuable suggestions.
References
[MG90] MacEachren, A. M., & Ganter, J. H. (1990). A
PATTERN
IDENTIFICATION
APPROACH
TO
CARTOGRAPHIC VISUALIZATION. Cartographica: The
International Journal for Geographic Information and
Geovisualization, 27(2), 6481. http://doi.org/10.3138/M2261337-2387-3007
[Als14] Alsallakh, B. (2014). A Visual Analytics Approach to
Segmenting and Labeling Multivariate Time Series Data.
[AAW07] Andrienko, G., Andrienko, N., & Wrobel, S. (2007).
Visual Analytics Tools for Analysis of Movement Data. SIGKDD
Explor.
Newsl.,
9(2),
3846.
http://doi.org/10.1145/1345448.1345455
[AAG00] Andrienko, N., Andrienko, G., & Gatalsky, P. (2000).
Visualization of spatio-temporal information in the Internet. In
11th International Workshop on Database and Expert Systems
Applications, 2000. Proceedings (pp.577585).
[AAG03] Andrienko, N., Andrienko, G., & Gatalsky, P. (2003).
Exploratory spatio-temporal visualization: an analytical review.
Journal of Visual Languages & Computing, 14(6), 503541.
[APP11] Archambault, D., Purchase, H. C., & Pinaud, B. (2011).
Animation, Small Multiples, and the Effect of Mental Map
Preservation in Dynamic Graphs. IEEE Transactions on
Visualization and Computer Graphics, 17(4), 539552.
http://doi.org/10.1109/TVCG.2010.78
[Ber10] Bertin, J. (2010). Semiology of Graphics: Diagrams,
Networks, Maps (1 edition). Redlands, Calif: Esri Press.
[CK98] Carlis, J. V., & Konstan, J. A. (1998). Interactive
Visualization of Serial Periodic Data.
[CKSS*14] Clark, K., Korth, K., Smith, L., Somes, R., Valent,
M., Winkler, P., & Henderson, B. (2014). New Jersey Bald Eagle
Project.
[DG14] Do, T. M. T., & Gatica-Perez, D. (2014). The Places of
Our Lives: Visiting Patterns and Automatic Labeling from
Longitudinal Smartphone Data. IEEE Transactions on Mobile

1
Computing,
13(3), 638648. http://doi.org/10.1109/TMC.2013.19
0
[DH02] Dragicevic, P., & Huot, S. (2002). SpiraClock: A Continuous
and Non-Intrusive Display for Upcoming Events. In IN: EXTENDED
ABSTRACTS OF CHI02 (pp. 604605).
[FQ11] Farrugia, M., & Quigley, A. (2011). Effective Temporal Graph
Layout: A Comparative Study of Animation Versus Static Display
Methods. Information Visualization, 10(1), 4764.
[FFM12] Fischer, F., Fuchs, J., & Mansmann, F. (2012). Clockmap:
Enhancing circular treemaps with temporal glyphs for time-series data.
In Proc. EuroVis Short Papers (pp. 97101).
[FNPS99] Fredrikson, A., North, C., Plaisant, C., & Shneiderman, B.
(1999). Temporal, Geographical and Categorical Aggregations Viewed
Through Coordinated Displays: A Case Study with Highway Incident
Data. In Proceedings of the 1999 Workshop on New Paradigms in
Information Visualization and Manipulation in Conjunction with the
Eighth ACM Internation Conference on Information and Knowledge
Management (pp. 2634). New York, NY, USA: ACM.
[FFMB*13] Fuchs, J., Fischer, F., Mansmann, F., Bertini, E., &
Isenberg, P. (2013). Evaluation of Alternative Glyph Designs for Time
Series Data in a Small Multiple Setting. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems (pp. 32373246).
New York, NY, USA: ACM. http://doi.org/10.1145/2470654.2466443
[GXYZ*14] Guo, C., Xu, S., Yu, J., Zhang, H., Wang, Q., Xia, J.,
Ebert, D. (2014). Dodeca-rings map: Interactively finding patterns and
events in large geo-temporal data. In 2014 IEEE Conference on Visual
Analytics Science and Technology (VAST) (pp. 353354).
[GCML06] Guo, D., Chen, J., MacEachren, A. M., & Liao, K. (2006).
A visualization system for space-time and multivariate patterns (visstamp). Visualization and Computer Graphics, IEEE Transactions on,
12(6), 14611474.
[Hg70] Hgerstrand, T. (1970). What about people in Regional
Science? Papers of the Regional Science Association, 24(1), 621.
http://doi.org/10.1007/BF01936872
[HDKS05] Hao, M. C., Dayal, U., Keim, D. A., & Schreck, T. (2005).
Importance-driven visualization layouts for large time series data. In
IEEE Symposium on Information Visualization, 2005. INFOVIS 2005
(pp. 203210). http://doi.org/10.1109/INFVIS.2005.1532148
[HHN00] Havre, S., Hetzler, B., & Nowell, L. (2000). ThemeRiver:
Visualizing Theme Changes over Time. In Proc. IEEE Symposium on
Information Visualization (pp. 115123).
[HS12] Heer, J., & Shneiderman, B. (2012). Interactive Dynamics for
Visual Analysis. Commun. ACM, 55(4), 4554.
[HEH09] Huang, W., Eades, P., & Hong, S.-H. (2009). Measuring
Effectiveness of Graph Visualizations: A Cognitive Load Perspective.
Information Visualization, 8(3), 139152.
[KW04] Kapler, T., & Wright, W. (2004). GeoTime information
visualization. In IEEE Symposium on Information Visualization, 2004.
INFOVIS 2004 (pp. 2532). http://doi.org/10.1109/INFVIS.2004.27
[Kar94] Karam, G. M. (1994). Visualization Using Timelines. In
Proceedings of the 1994 ACM SIGSOFT International Symposium on
Software Testing and Analysis (pp. 125137). New York, NY, USA:
ACM. http://doi.org/10.1145/186258.187157
[Kei00] Keim, D. A. (2000). Designing pixel-oriented visualization
techniques: Theory and applications. Visualization and Computer
Graphics, IEEE Transactions on, 6(1), 5978.
[KKA95] Keim, D. A., Kriegel, H.-P., & Ankerst, M. (1995).
Recursive pattern: a technique for visualizing very large amounts of
data. In , IEEE Conference on Visualization, 1995. Visualization 95.
Proceedings (pp. 279286, 463).
[KSS04] Keim, D. A., Schneidewind, J., & Sips, M. (2004).
CircleView: A New Approach for Visualizing Time-related
Multidimensional Data Sets. In Proceedings of the Working Conference
on Advanced Visual Interfaces (pp. 179182). New York, NY, USA:
ACM. http://doi.org/10.1145/989863.989891
[NSMP11] Noulas, A., Scellato, S., Mascolo, C., & Pontil, M. (2011).
An Empirical Study of Geographic User Activity Patterns in Foursquare.
ICwSM, 11, 70-573.
[PKMG*07] Pelekis, N., Kopanakis, I., Marketos, G., Ntoutsi, I.,
Andrienko, G., & Theodoridis, Y. (2007). Similarity Search in
Trajectory Databases. In 14th International Symposium on Temporal
Representation and Reasoning (pp. 129140).
[PVF13] Perin, C., Vuillemot, R., & Fekete, J.-D. (2013).
SoccerStories: A Kick-off for Visual Soccer Analysis. IEEE
Transactions on Visualization and Computer Graphics, 19(12), 2506
2515. http://doi.org/10.1109/TVCG.2013.192

[Peu94] Peuquet, D. J. (1994). Its About Time: A Conceptual

Framework for the Representation of Temporal Dynamics in
Geographic Information Systems. Annals of the Association of
American Geographers, 84(3), 441461.
[PMRW*96] Plaisant, C., Milash, B., Rose, A., Widoff, S., &
Shneiderman, B. (1996). LifeLines: Visualizing Personal
Histories. In Proceedings of the SIGCHI Conference on Human
Factors in Computing Systems (pp. 221227). New York, NY,
USA: ACM. http://doi.org/10.1145/238386.238493
[RPNG*08] Rinzivillo, S., Pedreschi, D., Nanni, M., Giannotti,
F., Andrienko, N., & Andrienko, G. (2008). Visually Driven
Analysis of Movement Data by Progressive Clustering.
Information Visualization, 7(3), 225239.
[RFFL08] Robertson, G., Fernandez, R., Fisher, D., Lee, B., &
Stasko, J. (2008). Effectiveness of Animation in Trend
Visualization. IEEE Transactions on Visualization and Computer
Graphics, 14(6), 13251332.
[SWWH11] Scheepens, R., Willems, N., van de Wetering, H., &
Van Wijk, J. J. (2011, March). Interactive visualization of
multivariate trajectory data with density maps. In Pacific
Visualization Symposium (PacificVis), 2011 IEEE (pp. 147-154).
IEEE.
[SH10] Segel, E., & Heer, J. (2010). Narrative Visualization:
Telling Stories with Data. IEEE Transactions on Visualization and
Computer Graphics, 16(6), 11391148.
[Shn96] Shneiderman, B. (1996). The eyes have it: a task by data
type taxonomy for information visualizations. In , IEEE
Symposium on Visual Languages, 1996. Proceedings (pp. 336
343). http://doi.org/10.1109/VL.1996.545307
[SMZZ13] Shrestha, A., Miller, B., Zhu, Y., & Zhao, Y. (2013).
Storygraph: Extracting Patterns from Spatio-temporal Data. In
Proceedings of the ACM SIGKDD Workshop on Interactive Data
Exploration and Analytics (pp. 95103). New York, NY, USA:
ACM. http://doi.org/10.1145/2501511.2501525
[SDWC07] Slingsby, A., Dykes, J., Wood, J., & Clarke, K.
(2007, July). Interactive tag maps and tag clouds for the
multiscale exploration of large spatio-temporal datasets. In
Information Visualization, 2007. IV'07. 11th International
Conference (pp. 497-504). IEEE.
[TM12] Tanahashi, Y., & Ma, K.-L. (2012). Design
Considerations for Optimizing Storyline Visualizations. IEEE
Transactions on Visualization and Computer Graphics, 18(12),
26792688. http://doi.org/10.1109/TVCG.2012.212
[TA15] Tominski, C., & Aigner, W. (n.d.). TimeViz Browser.
Retrieved March 30, 2015, from http://survey.timeviz.net/
[TGm83] Tufte, E. R., & Graves-Morris, P. R. (1983). The visual
display of quantitative information. Graphics press Cheshire, CT.
[Vas97] Vasiliev, I. (1997). Mapping Time. Cartographica: The
International Journal for Geographic Information and
Geovisualization, 34(2), 151. http://doi.org/10.3138/D357-234G2M62-4373
[WG12] Wongsuphasawat, K., & Gotz, D. (2012). Exploring
Flow, Factors, and Outcomes of Temporal Event Sequences with
the Outflow Visualization. IEEE Transactions on Visualization
and Computer Graphics, 18(12), 26592668.
[WGPW11] Wongsuphasawat, K., Guerra Gmez, J. A.,
Plaisant, C., Wang, T. D., Taieb-Maimon, M., & Shneiderman, B.
(2011). LifeFlow: Visualizing an Overview of Event Sequences. In
Proceedings of the SIGCHI Conference on Human Factors in
Computing Systems (pp. 17471756). New York, NY, USA:
ACM. http://doi.org/10.1145/1978942.1979196
[WDSC07] Wood, J., Dykes, J., Slingsby, A., & Clarke, K.
(2007). Interactive visual exploration of a large spatio-temporal
dataset: reflections on a geovisualization mashup. IEEE
Transactions on Visualization and Computer Graphics, 13(6),
11761183. http://doi.org/10.1109/TVCG.2007.70570
[Xiu11] Xiuli, Z. (2011). Progressive refinement for clustering
spatio-temporal semantic trajectories. In 2011 International
Conference on Computer Science and Network Technology
(ICCSNT) (Vol. 4, pp. 26952699).
[YKBN14] Yadav, K., Kumar, A., Bharati, A., & Naik, V.
(2014). Characterizing mobility patterns of people in developing
countries using their mobile phone data. In 2014 Sixth
International Conference on Communication Systems and
Networks (COMSNETS) (pp. 18).