PGX.UI: Visual Construction and Exploration of Large Property Graphs

Julia Kindelsberger

1,2,3,4

, Daniel Langerenken

1,2,3,4

, Malte Husmann

Korbinian Schmid

and Hassan Chaﬁ

Oracle Labs, 501 Island Parkway, 94002, Belmont, California, U.S.A.

Technische Universit

at M

unchen, Munich, Germany

Ludwig-Maximilians-Universit

at M

unchen, Munich, Germany

Universit

at Augsburg, Augsburg, Germany

Keywords:

Information Visualization, Graph Construction, Graph Drawing, Graph Exploration, Large Graph

Visualization, Property Graph, Summary Visualization, Graph Construction Time Line.

Abstract:

Transforming existing data into graph formats and visualizing large graphs in a comprehensible way are two

key areas of interest of information visualization. Addressing these issues requires new visualization ap-

proaches for large graphs that support users with graph construction and exploration. In addition, graph

visualization is becoming more important for existing graph processing systems, which are often based on the

property graph model. Therefore this paper presents concepts for visually constructing property graphs from

data sources and a summary visualization for large property graphs. Furthermore, we introduce the concept of

a graph construction time line that keeps track of changes and provides branching and merging, in a version

control like fashion. Finally, we present a tool that visually guides users through the graph construction and

exploration process.

1 INTRODUCTION AND

RELATED WORK

Graphs are a convenient and effective way for visu-

alizing relationships between data entities. By repre-

senting data entities as vertices and their ﬁne-grained

interconnections as edges, the graph visualization al-

lows users a quick and intuitive understanding of a

dataset. Graph processing systems such as graph

databases and graph analytic frameworks are gaining

popularity due to graph data from complex networks

(Newman, 2003) such as social networks (Hanne-

man and Riddle, 2005), biological networks (Olken,

2003), information networks (Kumar et al., 2000) and

technological networks (Shekhar et al., 1997). There

are several ways to model graphs as explained in pre-

vious research (Angles and Gutierrez, 2008). Our ap-

proach is based on the property graph model (Tinker-

pop, 2016) which recently has been used by multi-

ple popular graph databases (Kaliyar, 2015; Sevenich

et al., 2016; Neo4j, 2016; InﬁniteGraph, 2016; Titan,

2016). A property graph contains connected entities

(vertices) which can hold any number of properties

(key-value-pairs). Vertices are additionally tagged

with labels, representing different roles. Formally, a

property graph is a directed, binary, attributed multi-

graph. For example the vertex with the label Person

could have the properties Age and First Name.

Traditional large graph visualization techniques

include ﬁsh eye views (Sarkar and Brown, 1992),

clustering (Itoh and Klein, 2015) and highlights

(Liang and Huang, 2010). A more recent approach

is StructMatrix (Gualdron et al., 2015), which is a

methodology aimed at high-scalable visual inspec-

tion of graph structures with the goal of revealing

macro patterns of interest. Another work presents a

general-purpose framework for creating high-quality

edge bundlings from very large graphs (van der Zwan

et al., 2016). Yang et al. (2011) propose a summary

graph for summarizing relational database schemas.

Furthermore, research suggests 3D instead of 2D

graph visualization (Kwon et al., 2016). Moreover,

motion highlighting is suggested to support users in

comprehending large graph visualizations (Ware and

Bobrow, 2004).

At some point occlusion and data density are

too high for an effective visualization, the vertices

and edges overlap extensively and the user is over-

whelmed by the volume of information (Ware and

Bobrow, 2004). Therefore, it is commonly acknowl-

Kindelsberger J., Langerenken D., Husmann M., Schmid K. and Chaﬁ H.

PGX.UI: Visual Construction and Exploration of Large Property Graphs.

DOI: 10.5220/0006231603050310

In Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2017), pages 305-310

ISBN: 978-989-758-228-8

305

edged that graph visualizations with more than ﬁfty

vertices (hereinafter referred to as threshold) are un-

likely to be readily comprehended by the user with-

out interaction techniques (Ware and Bobrow, 2004).

Hence, previous work addresses various interaction

techniques (Herman et al., 2000) and visualization

concepts such as highlighting (Liang and Huang,

2010) and clustering (Itoh and Klein, 2015). Con-

sequently, tools were proposed that implement these

concepts. Jian Zhao et al. (2016) present a visualiza-

tion for meta-analysis based on user-authored annota-

tions. GraphVizdb (Bikakis et al., 2016) is a platform

that allows the user to interact with graphs on multiple

layers where the dataset can be explored at different

levels of granularity.

Finally, there is research that deals with graph

construction (Lequay et al., 2015) and research that

includes time aspects to graphs that change over time

(Valdivia et al., 2015).

However, none of these solutions combines vi-

sual graph construction, a graph summary visualiza-

tion and a time line. Therefore, this paper presents

a visualization approach for constructing a summary

visualization of large property graphs with property

previews and a tool that implements these concepts.

In contrast to other proposed summary visualizations,

our summary visualization is based on the property

graph model and focuses on property previews and

ﬁltering. Previews are based on the data type or the

number of occurrences. With this summary visualiza-

tion and the previews, the graph information is sum-

marized and shown to the user in a comprehensive

way. Filters can be applied to properties to create a

subgraph and the time line helps to navigate back and

forth in the graph construction process.

The contributions of our work are:

• A visualization concept to construct property

graphs from data sources.

• A summary visualization concept for large prop-

erty graphs.

• A concept for a property graph construction time

line.

• A tool for the graph construction and exploration

process.

The remainder of this paper is structured as fol-

lows: The proposed visualization approach is de-

scribed in section 2, which consists of the data ﬂow

of our proposal, the visual property graph construc-

tion, the summary visualization and the construction

time line. Section 3 concludes this paper by giving a

short summary and providing topics for future work.

2 PROPOSED VISUALIZATION

The aim of the proposed visualization is to provide

an intuitive way to visually construct property graphs

from data sources and help the user to discover large

graphs and hide unnecessary data. The result is a

summary visualization for large property graphs. Pre-

views of properties provide an overview of the infor-

mation contained in the graph. After applying ﬁl-

ters to reduce the number of vertices and edges, the

subgraph is visualized and can be further explored.

A time line keeps track of changes and provides ad-

vanced concepts such as branching and merging.

2.1 Dataﬂow

This section brieﬂy explains the dataﬂow of the visual

construction and exploration of large property graphs.

The ﬁrst step is to import the data sources. We refer

to data sources as any type of input data (e.g. CSV

ﬁles or databases). The second step is to show a pre-

view of the data sources to help the user to understand

the data. In addition, the tool provides a way to select

only speciﬁc data. Columns can be excluded and the

values can be ﬁltered. As a third step the data sources

are loaded visually, represented by table-like entities

that show the name of the data source and the asso-

ciate properties with their data type. The fourth step

is to connect the entities by dragging an entity onto

another entity. After establishing all connections, the

ﬁfth step is to apply ﬁlters to properties and gener-

ate the summary visualization. The sixth step is to

further ﬁlter the properties until the subgraph is small

enough to be displayed. Finally, this subgraph can be

explored by expanding vertices. I.e. users can click

on the vertex and all vertices connected to that vertex

appear. If the number of connected vertices is higher

than the threshold a single bundled vertex appears that

indicates the number of connected vertices. The prop-

erties of the bundled vertex can then be ﬁltered again.

2.2 Visual Property Graph

Construction

Traditionally, the graph construction has to be done

manually by the user with a code-based approach.

Our tool supports the user in the graph construction

process by visualizing the content of the data sources

and suggesting connections based on the data. Real-

world data is often not suited for visualizing graphs

without prior processing. Therefore, the ﬁrst step is to

transform data sources into a graph accepted format.

Often, data sources come from relational databases.

Before visualizing the data sources a preview of the

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

306

content is shown to the user and data can be excluded.

An example would be to deselect speciﬁc columns

in table-like data sources. Furthermore, properties

can be ﬁltered initially (e.g. specifying property Age

> 18). After the initial ﬁltering of values the data

sources are loaded visually and the user can create

connections via drag and drop. An example for visu-

ally loaded entities is shown in Figure 1.

Figure 1: User Interface of Loaded Entities.

In this way, connections between the same data

type or other matches are constructed. The user

can choose from the matches that the tool proposes

based on the data type and the number of occurrences.

These connections are represented as an edge in the

resulting summarization graph. Another type of con-

nection results from entities that contain information

on how other data sources are connected. This is

prominent in the area of relational databases and is

often referred to as junction tables. For instance, the

entity uses contains the ids of Customer and Product.

When connecting Customer via uses with Product the

uses entity visually collapses and is shown when hov-

ering over the edge between Customer and Product.

The thickness of the edge indicates that there are col-

lapsed entities. Similarly, the entity collapses into a

self-connection if two properties of the same entity

are connected. An example for the connected entities

is illustrated in Figure 2.

Figure 2: User Interface of Connected Entities.

2.3 Summary Visualization and

Exploration of Large Property

Graphs

Graphs often consist of thousands of vertices and

edges. Traditionally, large graph visualization meth-

ods focus on visualizing each vertex and edge and

may offer interaction techniques to hide parts of the

graph or summarize communities. We propose a sum-

mary visualization and exploration concept for large

property graphs that approaches the challenge from a

different angle. We visualize the summarization and

provide property previews to support the user in un-

derstanding the data.

2.3.1 Summary Visualization

The goal of our method is to visualize a large prop-

erty graph in an understandable way. For this reason,

we present a summary visualization that bundles ver-

tices and edges by labels. The size of the vertices and

the thickness of the edges correspond to the number

of edges and vertices with that label. Hence, the dif-

ferent roles of vertices and the connections are visu-

alized in a compact way. In addition, the properties

of each vertex role can be further explored. The pre-

views of the properties show the distribution of prop-

erties from all vertices with the same label. When the

number of vertices after applying the ﬁlters is below

a certain threshold the subgraph is visualized and can

be further explored. Figure 3 shows the summary

visualization of vertices with the labels Product, Em-

ployee and Customer and the edges with the labels

uses, works on, friend and sold.

Figure 3: User Interface of Summary Visualization for

Large Property Graphs.

2.3.2 Property Previews

A preview of the distribution of the properties is

shown for every vertex and edge. There are differ-

PGX.UI: Visual Construction and Exploration of Large Property Graphs

307

ent types of previews, depending on the data type of

the property and the number of occurrences. In this

section we propose a non-exhaustive list of previews

for different data types and give an example each. The

distribution of an integers property can be shown in a

bar chart. An example for this use case is the prop-

erty Age for the vertex with the label Employee. The

x-axis represents the age and the y-axis the number of

employees with a certain age. String properties can

also be shown in a bar chart. Another type of pre-

view displays a drop-down with common values with

their number of occurrences. Location properties (e.g.

property names lat and long) can be displayed on a

map. As an example, the colors of a speciﬁc state

indicates how many employees live in that speciﬁc

state. Pie charts can be used for properties with enum

characteristics. For instance, the vertex with the label

Product with the property Type is visualized as a pie

chart. Similarly, ﬂoat, boolean and id data types can

be visualized as line charts, switches and lists, respec-

tively. In this way the most ﬁtting preview is chosen

based on the data type and the number of occurrences.

This preview is chosen automatically, but the preview

can be changed by users manually. All previews sup-

port selections in an intuitive way. Hence, speciﬁc

values as well as ranges can be selected in order to ﬁl-

ter the properties of the summary visualization. Fig-

ure 4 illustrates examples for property previews.

Figure 4: User Interface of Property Previews.

2.3.3 Exploration

If the number of vertices, after applying ﬁlters on the

properties, is below the threshold the resulting sub-

graph is shown in addition to the summary graph. The

vertices of the subgraph now represent one speciﬁc

entity (e.g. an employee with a speciﬁc name and

age). The neighbours of each vertex can be expanded.

If the number of neighbours is above a certain thresh-

old a bundled vertex is visualized that indicates the

number of neighbours (e.g. a bundled vertex with the

label + 59 Friends if this employee has 59 edges with

the label friend to other employees). The subgraph

with a bundled vertex is shown in Figure 5.

Figure 5: User Interface of the Subgraph with a Bundled

Vertex.

The properties of this bundled vertex can be ﬁl-

tered in the same way as the properties of the vertices

of the summary visualization. If the number of ﬁl-

tered vertices is below the threshold the bundled ver-

tex disappears and the connected vertices appear. Fig-

ure 6 shows an overview of our tool, i.e. the summary

visualization, the subgraph and the time line.

Figure 6: Visualization Tool Overview: Showing (A) the

summary visualization, (B) the subgraph and (C) the time

line.

2.4 Property Graph Construction Time

Line

A time line keeps track of all changes in the graph

creation and exploration process visually. The x-axis

represents the time and the y-axis represents the cur-

rent number of vertices. Changes include, loading

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

308

data sources visually, making connections, ﬁltering

the data source, generating the summary graph, ﬁl-

tering the properties and generating and exploring the

subgraph. Various possibilities can be explored by

introducing the concepts of branching and merging

of time lines. An example of a time line with three

branches and one merging step is shown in Figure 7.

2.4.1 Branching

Branching a time line at a speciﬁc point copies the

applied ﬁlters and creates a new branch. All changes

made on that branch do not affect the original branch.

One branch is visualized at the time and the visualiza-

tion updates when the user switches to another point

of time or another branch. A branch can be created

from any point, and can be used to apply different ﬁl-

ters to the same starting point to explore the resulting

subgraph or the resulting number of vertices. In this

way, different paths of exploration can help the user

to compare changes. For instance, in one branch <18

is selected for the Age property of the vertex with the

label Employee and >18 in the other branch. In this

way, the graph can be compared between minors and

adults. Moreover, branches can be used for further

concepts such as copying branches for collaboration

and jumping back and forth to certain time points eas-

ily i. e. reverting or adding changes. The branch can

be switched at any time to jump back and forth to get

the corresponding visualization.

Figure 7: User Interface of the Time Line.

2.4.2 Merging

After exploring different branches, the ﬁlters of two

branches can be merged again. The semantics of the

merge can be speciﬁed manually as union, intersec-

tion or set difference. Union includes all applied ﬁl-

ters that are in the ﬁrst branch or the second branch.

Intersection includes all applied ﬁlters that are in both

branches and set difference includes all applied ﬁlters

that are in the ﬁrst but not in the second branch.

3 CONCLUSION AND FUTURE

WORK

In this paper we present the approach to visually con-

struct graphs based on data sources. We introduce

the summary visualization for large property graphs

and discuss ways to explore corresponding subgraphs.

The time line concept creates a link between the graph

construction and exploration phase.

Future work includes a user study to evaluate our

graph construction concept by comparing it to exist-

ing approaches. Moreover, we are going to evaluate

our summary visualization by applying it to graphs

with different structures such as graphs with different

numbers of vertex and edge labels. The goal is to un-

derstand the advantages and disadvantages of our con-

cepts for users with and without background in com-

puter science and/or graph visualization. We would

also like to understand how comprehensible our ap-

proach is in comparison to the clustering approach.

The goal is to let users explore graphs in both ways

and understand the differences and advantages of our

approach. The performance in completing tasks and

multiple choice questions about the comprehensibil-

ity are going to support our evaluation.

Furthermore, we are going to research Knowledge

Graphs such as DBpedia (Auer et al., 2007) and the

Google Knowledge Graph (Google, 2012) in the con-

text of our proposal. These are multi-relational, dense

graphs rich on meta data which consist of hundreds of

different labels and properties. Future work for dense

graphs could include machine learning to ﬁlter labels

for the user and making predictions based on property

names and values.

REFERENCES

Abello, J., Ham, F. V., and Krishnan, N. (2006). Ask-

graphview: A large scale graph visualization system.

IEEE Transactions on Visualization and Computer

Graphics, 12(5):669–676.

Angles, R. and Gutierrez, C. (2008). Survey of graph

database models. ACM Comput. Surv., 40(1):1:1–

1:39.

Auer, S., Bizer, C., Kobilarov, G., Lehmann, J., Cyga-

niak, R., and Ives, Z. (2007). Dbpedia: A nu-

cleus for a web of open data. In Proceedings of

the 6th International The Semantic Web and 2Nd

Asian Conference on Asian Semantic Web Conference,

ISWC’07/ASWC’07, pages 722–735, Berlin, Heidel-

berg. Springer-Verlag.

Bikakis, N., Liagouris, J., Krommyda, M., Papastefanatos,

G., and Sellis, T. (2016). graphvizdb: A scalable plat-

form for interactive large graph visualization. In 2016

PGX.UI: Visual Construction and Exploration of Large Property Graphs

309

IEEE 32nd International Conference on Data Engi-

neering (ICDE), pages 1342–1345.

Bollacker, K., Evans, C., Paritosh, P., Sturge, T., and Tay-

lor, J. (2008). Freebase: A collaboratively created

graph database for structuring human knowledge. In

Proceedings of the 2008 ACM SIGMOD International

Conference on Management of Data, SIGMOD ’08,

pages 1247–1250, New York, NY, USA. ACM.

Google (2012). Introducing the knowledge graph: things,

not strings. https://googleblog.blogspot.com/2012/05/

introducing-knowledge-graph-things-not.html. Ac-

cessed: 2016-11-15.

Gualdron, H., Cordeiro, R. L. F., and Rodrigues, J. F.

(2015). Structmatrix: Large-scale visualization of

graphs by means of structure detection and dense ma-

trices. In Proceedings of the 2015 IEEE International

Conference on Data Mining Workshop (ICDMW),

ICDMW ’15, pages 493–500, Washington, DC, USA.

IEEE Computer Society.

Hanneman, R. A. and Riddle, M. (2005). Introduction to

social network methods.

Herman, I., Melancon, G., and Marshall, M. S. (2000).

Graph visualization and navigation in information vi-

sualization: A survey. IEEE Transactions on Visual-

ization and Computer Graphics, 6(1):24–43.

InﬁniteGraph (2016). Inﬁnitegraph. http://www.objectivity.

com/products/inﬁnitegraph/. Accessed: 2016-11-10.

Itoh, T. and Klein, K. (2015). Key-node-separated graph

clustering and layouts for human relationship graph

visualization. IEEE Computer Graphics and Applica-

tions, 35(6):30–40.

Kaliyar, K. R. (2015). Graph databases: A survey. In Com-

puting, Communication Automation (ICCCA), 2015

International Conference on, pages 785–790.

Kumar, R., Raghavan, P., Rajagopalan, S., Sivakumar, D.,

Tompkins, A., and Upfal, E. (2000). The web as

a graph. In Proceedings of the Nineteenth ACM

SIGMOD-SIGACT-SIGART Symposium on Principles

of Database Systems, PODS ’00, pages 1–10, New

York, NY, USA. ACM.

Kwon, O. H., Muelder, C., Lee, K., and Ma, K. L. (2016). A

study of layout, rendering, and interaction methods for

immersive graph visualization. IEEE Transactions on

Visualization and Computer Graphics, 22(7):1802–

1815.

Lequay, V., Ringot, A., Haddad, M., Effantin, B., and Khed-

douci, H. (2015). Graphexploiter: Creation, visual-

ization and algorithms on graphs. In Proceedings of

the 2015 IEEE/ACM International Conference on Ad-

vances in Social Networks Analysis and Mining 2015,

ASONAM ’15, pages 765–767, New York, NY, USA.

ACM.

Liang, J. and Huang, M. L. (2010). Highlighting in infor-

mation visualization: A survey. In 2010 14th Inter-

national Conference Information Visualisation, pages

79–85.

Neo4j (2016). Neo4j graph database. https://neo4j.com/.

Accessed: 2016-11-10.

Newman, M. E. (2003). The structure and function of com-

plex networks. SIAM review, 45(2):167–256.

Olken, F. (2003). Tutorial on graph data management for

biology. In IEEE Computer Society Bioinformatics

Conference (CSB), volume 3.

Sarkar, M. and Brown, M. H. (1992). Graphical ﬁsh-

eye views of graphs. In Proceedings of the SIGCHI

Conference on Human Factors in Computing Systems,

CHI ’92, pages 83–91, New York, NY, USA. ACM.

Sevenich, M., Hong, S., van Rest, O., Wu, Z., Banerjee,

J., and Chaﬁ, H. (2016). Using domain-speciﬁc lan-

guages for analytic graph databases. Proc. VLDB En-

dow., 9(13):1257–1268.

Shekhar, S., Coyle, M., Goyal, B., Liu, D.-R., and Sarkar,

S. (1997). Data models in geographic information sys-

tems. Commun. ACM, 40(4):103–111.

Suchanek, F. M., Kasneci, G., and Weikum, G. (2007).

Yago: A core of semantic knowledge. In Proceed-

ings of the 16th International Conference on World

Wide Web, WWW ’07, pages 697–706, New York,

NY, USA. ACM.

Tinkerpop (2016). Deﬁning the property graph model.

http://tinkerpop.apache.org/docs/current/reference/.

Accessed: 2016-11-10.

Titan (2016). Titan distributed graph database. http://titan.

thinkaurelius.com/. Accessed: 2016-11-10.

Valdivia, P., Dias, F., Petronetto, F., Silva, C. T., and

Nonato, L. G. (2015). Wavelet-based visualization of

time-varying data on graphs. In Visual Analytics Sci-

ence and Technology (VAST), 2015 IEEE Conference

on, pages 1–8.

van der Zwan, M., Codreanu, V., and Telea, A. (2016).

Cubu: Universal real-time bundling for large graphs.

IEEE Transactions on Visualization and Computer

Graphics, 22(12):2550–2563.

Ware, C. and Bobrow, R. (2004). Motion to support

rapid interactive queries on node–link diagrams. ACM

Trans. Appl. Percept., 1(1):3–18.

Yang, X., Procopiuc, C. M., and Srivastava, D. (2011).

Summary graphs for relational database schemas.

Zhao, J., Glueck, M., Breslav, S., Chevalier, F., and Khan,

A. (2016). Annotation graphs: A graph-based vi-

sualization for meta-analysis of data based on user-

authored annotations. IEEE Transactions on Visual-

ization and Computer Graphics, PP(99):1–1.

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

310