Inside Mall: Visual Analytics of Customer Behavior and Activities

Jens Opolka

, Patrick Riehmann

, Heiko Peter

and Bernd Froehlich

Virtual Reality and Visualization Research Group, Bauhaus-Universit

at Weimar, Bauhausstraße 11, Weimar, Germany

Favendo GmbH, Goethestraße 3b, Jena, Germany

Fraunhofer IAIS, Schloss Birlinghoven, Germany

Keywords:

Embedded Flow Visualization, Customer Behavior Visualization, Customer Tracking, Floor Plan Visualiza-

tion, Indoor Map Visualization.

Abstract:

This paper presents a coordinated multi-view system consisting of visualizations for displaying customer be-

havior and activities in a shopping center environment based on indoor tracking information gathered by Blue-

tooth beacons. Different perspectives to ﬁnd structures and hidden patterns within the data set are supported

including different customer ﬂow visualization methods based on actual ﬂoor plans as well as abstract ﬂow

graphs. These are linked and coordinated with Parallel Sets to gain insight into the speciﬁcs of the customer

base and with different time-oriented visualizations to show and to compare the different periods of customer

presence.

1 INTRODUCTION

The rise of the Bluetooth beacon technology has dra-

matically evolved indoor tracking. This cheap tech-

nology equips shopping centers with the ability to

track customer smartphones throughout the mall and

to enable location-based services indoors via smart-

phones. The data collected by the beacons, as well

as by the accompanying smartphone app, consists of

thousands of customers and hundreds of thousands of

customer events. The customer events reﬂect the in-

teraction of an individual with a shopping center. Size

and complexity of the data present a challenge to mar-

ket research. Furthermore, there are different parties

with various interests to consider. For instance, a store

owner (in a mall) might be interested in the activity of

his own store during a speciﬁc time of day. The man-

ager of a mall, on the other hand, might be more in-

terested in the pathing of all customers throughout the

mall. In the discussion with our industry partner, we

gathered a set of typical tasks and related information

needs such as

• Location planning for venues in order to draw cus-

tomers to less active areas: which areas of a ﬂoor

are the most active?

• Positional planning for advertisements (e.g. bill-

boards): what are the most prominent routes

through the shopping center?

• Identiﬁcation of competing businesses or poten-

tial partners for cooperation: which venues attract

an equal group of customers?

• Staff and stock planning: when is the most active

time or season for a venue?

• Identiﬁcation of potential times to run a campaign

in order to increase the number of customers:

when is the least active time or season?

• Creation of personalized offers for spe-

ciﬁc customer groups: what are the

speciﬁcs/characteristics of the customer base

of a venue?

• Evaluation of the success of campaigns: do

behavioral patterns exist for certain customer

groups?

• Identiﬁcation of the possible need to improve cus-

tomer loyalty: how many customers visit a venue

more than once?

Our coordinated multi-view system for analyzing

customer behavior contributes genuine ﬂow visualiza-

tions, particularly designed for different analytic sce-

narios, either embedded in ﬂoor plans as a directed

ﬂow visualization for expressing consumer ﬂows be-

tween selected shops or as routed ﬂow visualizations

for investigating consumer trafﬁc of multiple venues

in a vicinity or on an entire mall ﬂoor.

Additionally, we utilize abstract (non-embedded)

ﬂow displays for ﬁnding relationships between shops

Opolka, J., Riehmann, P., Peter, H. and Froehlich, B.

Inside Mall: Visual Analytics of Customer Behavior and Activities.

DOI: 10.5220/0007284901830190

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 183-190

ISBN: 978-989-758-354-4

183

that are neither adjacent nor in the same vicinity, but

share the same consumer base. These novel visual-

izations are accompanied by established ones such

as Parallel Sets and calendar views for covering set-

based and temporal aspects.

2 RELATED WORK

Customer tracking is important in retail analysis

(Blattberg and Deighton, 1991). The ubiquity of

smart phones made tracking of individuals cheaper.

Despite that, indoor ﬂow visualization has only been

discussed occasionally in a scientiﬁc context. Moi-

seeva (Moiseeva and Timmermans, 2010) gave a good

overview of approaches used for pedestrian model-

ing and the technologies used for data gathering. A

previous method for monitoring indoor movement of

pedestrians (Liebig and Xu, 2012) used special sen-

sors to capture pedestrian quantities, frequencies, and

paths. Cell-Phones (You et al., 2011) are an alterna-

tive for tracking shopping activities as well as the ac-

tual time spent shopping and the money thereby spent.

These efforts were expanded (You et al., 2014) for

collecting and analyzing data for entire retail trade ar-

eas. Chao et al. (Chao et al., 2013) speciﬁcally focus

on consumer ﬂows embedded in a mall ﬂoor plan but

only with a limited number of stores (nine) with ar-

tiﬁcially generated data. A heatmap is offered but it

needs a switch between the two views. Berry (Berry,

2007) applied common approaches from GIS (geo-

graphic information system) to a single store by visu-

alizing movement patterns regarding spatial locations

of different product categories also with a heatmap.

Traditional ﬂow map visualizations as presented

by Tobler in 1987 (Tobler, 1987) or more recently by

Phan et al. (Phan et al., 2005) show the outgo-

ing/incoming ﬂow of a resource or a group of people

from/to a place of origin encoded as thickness of the

ﬂow line, but only for simple and direct routings.

3 THE DATA SET

The data set was provided by our industry partner.

It contains two parts: First, the static, non-temporal

part of the data set that includes all objects needed

to model the shopping center itself and its customers;

nearly 200 different venues (stores, eateries, and fa-

cilities) and more than 170.000 customers. The sec-

ond part consists of the event database of all dynamic

aspects. It contains all customer related events which

represent customer activity connected to the shopping

center as well as the mobile app. It can be considered

as a stream of time-series data (timestamps) produced

by the user of the app (the actual customers) as well

as by the indoor beacons.

Besides the static/dynamic parts, we differentiate

(as orthogonal dimension) location-based data (e.g.

venues or venue related customer events) and non-

spatial objects (e.g. customers or categories). A

location-based object always includes speciﬁc coor-

dinates (latitude and longitude). Furthermore, the

data set contains several different kinds of object at-

tributes. Here we distinguish between continuous

data (e.g. timestamps), discrete, categorical data (e.g.

gender), and pseudo categorical / quantized data ( e.g.

age).

Figure 1: InsideMall with ﬁlter/option pane (left), the view

area (center) and info/selection pane (right).

4 GENERAL USER INTERFACE

The interface consists of the main area in the center,

the ﬁlter menu on the left, and the information pane to

the right (see Figure 1). The main area features a co-

ordinated multi-view setup in order to create multiple

views for looking at and comparing different visual-

izations and settings (e.g. different time windows) si-

multaneously. Each view can be split horizontally or

vertically to create a new viewport. A controlbar aids

in splitting, closing, and linking a view or switching

between the different visualizations.

On the left side, several ﬁlter options are offered,

mostly as scented widgets (Willett et al., 2007). For

instance, the scented histogram sliders are used to

restrict temporal maximum and minimum values in-

stantly in order to ﬁlter events in all visualizations

sensitive to it. Thus, we can observe and restrict the

temporal distributions at different granularity; for in-

stance, at day level (a Gaussian distribution with its

peak at 3 o’clock) and week level (weekends contain

more events than the work days).

The customer ﬁlters beneath are used to ﬁlter cus-

tomers by gender, mobile device, and age. Exﬁltrat-

ing a customer will also ﬁlter out events triggered by

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

184

that customer. Additionally, we offer ﬁlters for the

different categories and venue types.

On the right is a typical info/selection pane con-

sisting of an info box to show additional information

for selected objects (description and frequencies of

customers for a venue) on top and a venue list used

to select different venues beneath it. The venue list

shows all stores and eateries. Venues of different cat-

egories and ﬂoors are colored differently to represent

the type of business. Selections in the list inﬂuence

the active visualizations.

5 EMBEDDED FLOW

VISUALIZATION

The ﬂoor plan visualization (see Figure 2) is based on

the actual layout of the shopping center. The over-

all assumption is that the more customers enter the

area connected to a venue the more popular the venue

is. Of course one must consider appropriate dwell

times and time differences between the entry/exit area

events of persons simply passing through an area.

Stores are displayed in cyan and eateries in ma-

genta. Different data can be mapped on intensity (be-

ing sensitive to ﬁltering the events in the ﬁlter bar)

such as the total number of visits for a venue. This

gives an indication of the overall activity or the aver-

age number of visits per customer (the total number

of visits divided by the number of customers that ever

visited the store) which is related to customer loyalty.

The dwell time mode maps the total amount of time

customers spend in a zone within a certain time period

(adjustable in the left ﬁlter bar) or the average time a

customer spends in a venue. Additional labels show

the respective absolute numbers related to the current

mapping mode. Some interesting examples of typical

ﬁndings are shown in Figure 2.

In order to gain more insight into the movement

of customers (for instance, to follow a path of a group

of customers along a sequence of way points given

by the ﬂoor plan), a complex graph-like model with

one or many origins (the venues) is required. Some of

these way points may even be visited more than once

along the path, creating loops.

Direct Flow. In a shop-centered view the graph can

be separated with respect to the origin venues. Each

level of the subgraph may include nodes for each

venue in a shopping center. The level i of the sub-

graph shows all venues visited i venues after the ori-

gin venue (root node). The direct ﬂow visualizes the

outgoing (or incoming) ﬂow from a single venue to

Figure 2: The ﬂoor visualization can express the total num-

ber of visits (top left), average number of visits per customer

(top right), total dwell time (bottom left), and average dwell

time per customer (bottom right). The total number of visits

displays that the activity of customers is higher for venues

near the entrance of the ﬂoor to the right. However, the av-

erage visits of customers for each venue (bottom right) in-

dicates a better distribution of visits across the whole ﬂoor.

For the total dwell time of customers (bottom left), we can

see a similar distribution to the total number of visits. It’s

not surprising that venues that are visited more often also

show higher dwell times. Furthermore average dwell times

per customer (bottom right) show that customers tend to

spend more time in larger venues covering a bigger area.

We can also observe high dwell times for restaurants and

coffee shops.

the next (previous) venue. In Figure 3 we can see the

ﬁrst level of the outgoing ﬂow from a selected store

to all subsequent stores visited by different groups of

customers. The line thickness maps the number of

customer transitions between the given shop to one

of the next shops. This allows analysts to get an

overview of where customers that frequent a partic-

ular venue are most likely to go afterwards.

However, if we want to display more levels in or-

der to show the second, third, or subsequent venues

being visited after the initial venue, the visualiza-

tion becomes increasingly difﬁcult to read since the

straight ﬂow lines cover not only a lot of the ﬂoor

plan underneath but also produce a lot of crossings

with each other. This makes it very difﬁcult to follow

a speciﬁc path of the ﬂow.

Using color gradients for expressing the direction

improves the situation but if more than two levels of

the ﬂow layout are displayed even these gradients do

not help to overcome the visual problems caused by

the massive overplotting of the ﬂow lines.

Reducing the number of levels would mitigate the

Inside Mall: Visual Analytics of Customer Behavior and Activities

185

Figure 3: Direct outgoing ﬂow for a selected store, comparing the ﬂow for one level (left) and 20 level (right).

effects of overplotting but does not entirely solve the

problem. We offer several ﬁlters in the left ﬁlter pane

such as an area duration threshold ﬁlter. This de-

creases the number of events that contribute to the

ﬂow lines. We can also use a tree ﬁlter in order to ﬁl-

ter paths traversed by a limited number of customers

in favor of dominant ﬂow strands.

Routed Flow. A routed ﬂow aids in avoiding over-

plotting of the ﬂow lines and not obscuring the venues

in the ﬂoor plan by creating the ﬂow lines along a

predeﬁned path which follows the real corridors of a

shopping center and thus approximates the actual path

a customer would take (see Figure 4). An individual

path depicts a possible unique path through the shop-

ping center. The width or weight of an edge between

two nodes represents the number of customers who

have taken this path. The underlying graph structure

is created upon nodes for each store and additional

dummy nodes along the aisles (see Figure 5). The

number of customer events at consecutive venues are

encoded by the width of the edges .

Each edge of the graph is initialized with zero,

meaning that no customer has taken this path. In

order to update the edge weights, we iterate over

each customer mall visit and increase all edges of

the shortest paths between the consecutively visited

venues by one (considering appropriate dwell or du-

ration times at the venues). By doing so we create

a binning/summarization of the approximated paths

customers in the mall have taken which reveals the

predominant routes throughout the mall.

Pairwise Flow. The pairwise ﬂow layout was in-

spired (Chao et al., 2013) and shows a straight ﬂow

line for each pair of venues. The thicker a ﬂow line is

drawn, the more often customers have been in either

venue after visiting the other one (a transition from

one venue of a pair to the other exists). Even though

this seems to work very well in the setting described

in the original paper, it does not work in a bigger mall

setting where stores are not arranged strictly opposite

of each other (as in Chao’s paper).

Figure 4: Routed ﬂow for a selected store (top right).

One might argue that a curved drawing of the ﬂow

lines could enhance the legibility of this approach but

even then crossings cannot be avoided (see Figure 6

left). As with the direct ﬂow, reducing the number

of lines by ﬁltering helps (see Figure 6 right); for in-

stance, drawing only lines for pairs with a minimum

number of transitions or ﬁltering the events by the

time a customer spends in an area.

Comparison. The direct ﬂow layout shows the tran-

sitions from one selected venue to the next for a subset

of customers. Multiple customers with the same tran-

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

186

sition are bundled into one ﬂow line. For each origin

venue it is possible to follow the sequence of venues

visited by a group of customers.

Figure 5: Routed ﬂow with underlying graph structure.

Venue nodes are shown in red, dummy nodes are blue.

In the routed ﬂow variant, routes represent bins

that are incremented when a customer follows a spe-

ciﬁc path to another venue. Given the line width it

is quite obvious which routes/directions were mostly

taken after visiting a particular venue, but the infor-

mation about the actual sequence of venues cannot

be shown. For example, for an origin store it is no

longer possible to discern how many customers vis-

ited a target store directly after visiting the origin

venue compared to the number of customers who vis-

ited the target store two venues later. In the routed

ﬂow variant the overall path is the focus (and not the

order in which stores are traversed). Interestingly, the

quality/precision of the visualization increases with a

lower area duration threshold setting because, with a

lower threshold, more area events are taken into con-

sideration which allows for a better tracking of the

actual path taken by a customer. In contrast, for the

direct ﬂow variant where the direct transition between

stores is the focus and the actual path between two

stores is not important, the quality of the layout im-

proves with a higher area duration threshold because

the actual transition from one venue of interest (a

venue which is not only passed by a customer but ac-

tively visited) to the next venue of interest is shown.

Figure 6: Pairwise ﬂow unﬁltered (left) and ﬁltered showing

the 50 biggest transitions between venues (right).

6 ABSTRACT FLOW

VISUALIZATION

The previously described ﬂow methods aid in pre-

serving the mental map, but such an underlying plan

can be inﬂexible (for example, for showing the cus-

tomer ﬂow across different ﬂoors). For instance, what

attracts customers after entering the shopping center

or a ﬂoor? Was the venue approached directly after

entering or was it approached only after visiting sev-

eral other stores? Such detailed information across all

ﬂoors is very hard to come by in the ﬂoor plan based

variants.

We propose a layered ﬂow visualization where the

root node (focus node) for the origin venue is placed

at the center layer. To the right of the root node the

outgoing ﬂow is depicted whereas to the left the in-

coming ﬂow is shown. Each venue is represented as

a circle; its radius depending on the number of visits.

Edge thicknesses represent the number of customers

who visited the ﬁrst venue in the current layer and a

second venue in the next one. Crossing minimization

is done after (Sugiyama et al., 1981).

Figure 7: Flow graph for a store.

Figure 8: Aligned event chains for 3 customers (top). Each

color represents a different venue. Weighted ﬂow graph cre-

ated from event chains (bottom).

The possibility of showing both incoming and out-

going ﬂow for the focus venue at the same time is an-

other advantage over the ﬂoor plan ﬂow where it is

barely possible to show both ﬂow directions distinc-

tively (especially if color coding is already used to

express the node distance in the directed embedded

ﬂow, see Figure 3).

To create the non-embedded graph, ﬁrst we align

all customer event chains to the ﬁrst occurrence of the

Inside Mall: Visual Analytics of Customer Behavior and Activities

187

Figure 9: The ﬂow through a store on the third layer is highlighted.

focus venue (shown in yellow, see Figure 8). Across

all chains, events of the same venue are combined

into one node The weight of the resulting node corre-

sponds to the number of occurrences of events for that

venue in that level. Events prior to the focus node are

used to create the incoming ﬂow and following events

are used to create the outgoing one. The weight for an

edge connecting two nodes depends on the number of

equal consecutive event pairs with respective venues;

both at the correct layers.

The ﬂow can be restricted by numbers of layers

which are shown for the incoming and outgoing ﬂow,

by a maximum number of edges leaving a node, or

by a minimum number of events per node. The ini-

tial display is already restricted to emphasize the most

prominent ﬂows since most users are not interested in

a path only a few customers took.

Apart from those global ﬁlter options, direct in-

teraction is still crucial. The incoming and outgoing

ﬂow for any node highlighted by the mouse is traced

across all layers as previously proposed (Riehmann

et al., 2005) by highlighting the fraction of each edge

respective to the number of customers that also visited

the marked venue sometime in their event chain aside

from the focus venue (Figure 9). Double-clicking a

node will move it into focus and rebuild the graph.

A focus node can also be changed by selecting a

different venue in the sidepane or in the ﬂoor plan.

Horizontal zoom is implemented by increasing the

maximum number of layers shown at the same time.

Zooming in will decrease the number of layers shown

and vice versa.

7 PARALLEL SETS

The ﬂow visualizations focus on customer movement

and activity. Parallel Sets (Bendix et al., 2005) were

integrated for analyzing the categorical speciﬁcs of

customer base.

Figure 10: Parallel Sets comparing the customer base for

two separate stores (top row) and for the same store week-

days (bottom left) and weekends (bottom right). Store A

has more male than female customers (top axis) and two

major age groups: 21-40 and 41-60. The younger group

consists of more men than women, whereas the older group

is balanced. Store B, to the right, has slightly more female

than male customers and at a younger age. The third axis,

showing the mobile devices, may indicate a higher budget

(iPhone users tend to have a higher income (Hixon, 2014)).

On the bottom row we can see two visualizations compar-

ing two different time ranges for Store A; weekdays left and

weekends right. Gender distribution is similar but age dis-

tribution is not. Weekdays two-third of the customers are

under 40. On weekends the customers tend to be older,

more between 41 and 60 years of age. The younger group

also consists of signiﬁcantly more men than women as com-

pared to weekdays.

If no venue is selected, the visualization shows

the customer base for the whole shopping center by

default. Once a ﬂoor, store, or eatery has been se-

lected, the display is updated to show only those cus-

tomers who have visited the selected venues. This

way the customer base can be compared across dif-

ferent venues. An analyst can, for example, easily

use Parallel Sets to see whether the customer base of

a certain store includes more men or women and can

even compare these values for different time spans or

in comparison to other stores. The comparison of dif-

ferent sets of customers (e.g. for different stores) is

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

188

facilitated by the multi-view setup of the application.

This is simply done by opening a second viewport

showing the Parallel Set plot with the desired ﬁltering

applied. By using this method an arbitrary number of

settings can be compared to each other (only limited

by the available screen space). Figure 10 shows two

Parallel Sets (top row) comparing the customer base

for two separate stores from the category ’Fashion’.

8 EVENTS AND TIME

We depict user events in a calendar representation

which is easy to understand and interpret, even for

novice users. We use simple bar charts to indicate

the frequencies of customers (red) and events (cyan).

In Figure 11 (top) we can observe activity peaks be-

fore Christmas and for the time between Christmas

and New Year’s.

Figure 11: Overview row for end of December 2014,

showing increased activity around Christmas (top). Daily

overview for January 2015 (bottom).

The more detailed calendar view (see Figure 11,

bottom) consists of a row for each day subdivided by

the hours of the day. Weekends are highlighted in a

different color. We can see that the active times of the

mall range from about 8:00 to 20:00. Increased activ-

ity patterns around 18:00 can be observed on week-

ends while on weekdays 12:00 is the most active time.

Weekends are usually the most active days.

9 VIEW COORDINATION

Due to the variety of data types and their many differ-

ent aspects (time-dependent, location based, stream-

ing data, continuous- and categorical data) a single

visualization does not sufﬁce. Our coordinated multi-

view approach provides a personalized viewing setup

by creating multiple views of the same data set. By al-

lowing the user to link views together, interactions in

one viewport will automatically affect the other linked

viewports.

Figure 12: Multi-view setup showing the ﬂoor plan (left),

the ﬂow graph (upper right) and the parallel sets visualiza-

tion (bottom right).

Both are possible: exploring the same data con-

ﬁguration with different visualizations and comparing

different data using the same visualization. Thus we

introduced viewport-linking. For example, selecting a

store in the ﬂoor plan (or the venue list) should restrict

the presented data in the Parallel Sets visualization to

those customers who have visited that store, restrict

the events evaluated for the calendar visualization to

those events connected to that store, and set the focus

node for the ﬂow graph visualization to the same store

without having to activate each visualization individ-

ually and select the desired store manually. We enable

this functionality by linking any number of viewports

to the active viewport (dark control bar) by pressing

the chain-button in its control bar.

Once a viewport is linked, all performed interac-

tions in the active view are propagated to each linked

view. Views that are not linked keep their own set of

ﬁlters. In this regard we differentiate between high-

lighting and selecting an object. Highlighting an ob-

ject will only affect the current visualization. Select-

ing one or many objects (for example, a venue in the

ﬂoor plan or a node in the Flow Graph) will propagate

the selection to all linked views. If a venue is selected

this way, customers and events are ﬁltered to show a

sub set of the data for that venue.

10 CONCLUSION & FUTURE

WORK

In this paper we presented a coordinated multi-view

system aimed at analyzing customer behavior in an

indoor environment. The ﬂoor plan visualization pro-

vides overview of activity, attractiveness, and loy-

alty of customers. It aids the user’s mental map by

presenting the data in context of the spatial layout

Inside Mall: Visual Analytics of Customer Behavior and Activities

189

of the mall and showing the ﬂow from one store to

the next (direct ﬂow) or revealing predominant routes

(routed ﬂow) within the shopping center. The ab-

stract ﬂow graph offers a detailed analysis of the ﬂow

for non-adjacent venues. They are complemented

by a calendar view for revealing temporal customer

patterns and by Parallel Sets for analyzing customer

speciﬁcs. Our visualization system can be used by

non-expert users but still offers a lot of potential for

information extraction by domain experts.

More precise tracking data would entail a more

precise visualization with less need of route approx-

imation. Additional tracking within a venue would

further improve the resulting visualization. We see

potential for improving the appeal and legibility of

the ﬂoor plan ﬂow. Harmonizing the color schemes

and mappings across the different views would lead

to a more appealing visualization and might be more

comprehensible to the users. Furthermore, distorting

the ﬂoor plan to increase the space between venues

(the corridors) would reduce visual clutter without

changing the adjacencies of the venues and thus pre-

serve the known spatial context. The increased spaces

between venues also gives room to present more ad-

vanced ﬂow visualizations. Edge bundling of the ﬂow

lines could help regarding the legibility (for instance,

routings around obstacles (venues) or changes of the

ﬂow direction). For the direct ﬂow approach it would

help to reduce the areas obscured by the edges. How-

ever, this could hamper identifying the direct connec-

tions between shops.

The calendar views could be enriched to com-

pare patterns across months and seasons for different

years. We also propose a manual aggregation of time

spans by the user; for example by the possibility to

aggregate multiple weeks or days.

At a certain stage of development the question

about a 3D view arose from our industry partner. Nat-

urally, most malls do not consist of only a single ﬂoor.

While our abstract ﬂow visualizations can deal with

ﬂow across ﬂoors, a map-based solution to investigate

ﬂoor-to-ﬂoor statistics in detail remains future work.

ACKNOWLEDGEMENTS

This work was supported in part by the German Fed-

eral Ministry of Education and Research (BMBF) un-

der grant 03IP704 (project Intelligentes Lernen) and

grant 03IPT704X (project Big Data Analytics).

REFERENCES

Bendix, F., Kosara, R., and Hauser, H. (2005). Parallel sets:

Visual analysis of categorical data. In Information Vi-

sualization, 2005. INFOVIS 2005. IEEE Symposium

on, pages 133–140.

Berry, J. K. (2007). Map Analysis: Understanding Spatial

Patterns and Relationships. GeoTec Media.

Blattberg, R. C. and Deighton, J. (1991). Interactive mar-

keting: Exploiting the age of addressability. MITSloan

Management Review, 33(1).

Chao, T., Du, R., Gluck, J., Maidasani, H., and Wills,

K. (2013). C-ﬂow: Visualizing foot trafﬁc and

proﬁt data to make informative decisions. Tech-

nical report, University of Maryland. Online:

http://duruofei.com/Dev/c-ﬂow [last accessed January

2019].

Hixon, T. (2014). What kind of per-

son prefers an iphone? Online:

http://www.forbes.com/sites/toddhixon/2014/04/10/

what-kind-of-person-prefers-an-iphone/ [last ac-

cessed January 2019].

Liebig, T. and Xu, Z. (2012). Pedestrian monitoring system

for indoor billboard evaluation. Journal of Applied

Operational Research, 4:28–36.

Moiseeva, A. and Timmermans, H. (2010). Imputing rel-

evant information from multi-day gps tracers for re-

tail planning and management using data fusion and

context-sensitive learning. Journal of Retailing and

Consumer Services, 17(3):189–199.

Phan, D., Xiao, L., Yeh, R., Hanrahan, P., and Winograd, T.

(2005). Flow map layout. In Proceedings of the Pro-

ceedings of the 2005 IEEE Symposium on Information

Visualization, INFOVIS ’05, pages 219–224.

Riehmann, P., Hanﬂer, M., and Froehlich, B. (2005). Inter-

active sankey diagrams. In IEEE Symposium on In-

formation Visualization, 2005. INFOVIS 2005, pages

233–240.

Sugiyama, K., Tagawa, S., and Toda, M. (1981). Methods

for visual understanding of hierarchical system struc-

tures. Systems, Man and Cybernetics, IEEE Transac-

tions on, 11(2):109–125.

Tobler, W. R. (1987). Experiments in migration mapping by

computer. The American Cartographer, 14(2):155–

163.

Willett, W., Heer, J., and Agrawala, M. (2007). Scented

widgets: Improving navigation cues with embedded

visualizations. IEEE Transactions on Visualization

and Computer Graphics, 13(6):1129–1136.

You, C.-W., Kao, H.-L., Ho, B.-J., Chen, Y.-H., Wang, W.-

F., Bei, L.-T., Chu, H.-H., and Chen, M.-S. (2014).

Convenienceprobe: A phone-based system for re-

tail trade-area analysis. Pervasive Computing, IEEE,

13(1):64–71.

You, C.-W., Wei, C.-C., Chen, Y.-L., Chu, H.-H., and Chen,

M.-S. (2011). Using mobile phones to monitor shop-

ping time at physical stores. Pervasive Computing,

IEEE, 10(2):37–43.

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

190