Visualisation of RDF(S)-based Information

Visualisation of RDF(S)-based Information Alexandru Telea,Flavius Frasincar,and Geert-Jan Houben Eindhoven University of Technology

PO Box513,NL-5600MB Eindhoven,the Netherlands alext,?aviusf,houben@win.tue.nl

Abstract

As Resource Description Framework(RDF)reaches ma-turity,there is an increasing need for tools that support it.A common and natural representation for RDF data is a directed labeled graph.Although there are tools to edit and/or browse RDF graph representations,we found their architecture rigid and not easily amenable to pro-ducing effective visual representations,especially for large RDF graphs.We discuss here how GViz,a general pur-pose graph visualisation tool,allows the easy construction and?ne-tuning of various visual exploratory scenarios for RDF data.GViz’s extended ability of customizing the visu-alisation’s icons showed to be very useful in the context of RDF graph structures visualisation.Among the presented applications,we mention customizable selections,schema-instance comparison,instances comparison,and schemas comparison(schema evolution).GViz proved to be able not only to visualize large RDF data models,but also to be very ?exible in designing scenario-speci?c queries to support the exploration process.

1.Introduction

Resource Description Framework(RDF)is the web metadata language.It is used to describe information about web resources.The semantics associated with this infor-mation enables web applications interoperability.An RDF model that describes some web resources is also called an RDF instance.An RDFS schema can be used to de?ne ap-plication speci?c vocabularies.This schema can be associ-ated with an RDF instance in order to validate the instance. Both RDF instance and RDFS schema are RDF models.

As RDF reaches maturity,there is an increasing need of tools that allow users to understand,i.e.browse and mod-ify,RDF data.Two such tool types exist:textual and vi-sual.Examples of textual RDF browsing tools are Protege-2000[4],OntoEdit,and OntoMat.However versatile,ex-perience proved that analysis of moderately voluminous re-lational(graph)data is not effective in text based environ-ments[1].Therefore,in this paper we shall mainly focus on visual browsing tools.Examples of such tools special-ized for RDF browsing are IsaViz[5],FRODO RDFSViz, and OntoViz(visualisation plug-in for Protege-2000).In the following,we will discuss two of the above mentioned tools:Protege2000and IsaViz.

Protege-2000[4]is a textual browsing/editor tool for knowledge models.It enables modeling at conceptual level such that the user doesn’t need to be concerned with the syntax of the?nal output.One knowledge representation format supported by Protege-2000is RDF(S).Protege2000 uses the RDF API from Simple RDF Parser and Compiler (SiRPAC)for reading RDF models.For comparison pur-poses,we chose to represent the same newspaper example, the default Protege-2000project,in several browsing tools. Figure1depicts the newspaper RDFS schema in Protege-2000(without the visualisation plug-in).It is obvious that such text-based representations fail in conveying structural insight in anything but relatively small and simple RDF(S) datasets.A solution to this problem is the usage of a plug-in (i.e.OntoViz)that adds visualisation capabilities to the

tool.

Figure1.Newspaper schema in Protege

IsaViz[5]is a visual browsing tool for RDF models. IsaViz uses the RDF API of Jena for reading RDF mod-els and AT&T’s GraphViz package[3]for the graph lay-out.Some of the features that IsaViz supports are text-based search,copy and paste,model editing,editing of the visual shapes used for nodes and arcs,textual property browser, and graph/radar views(radar views open a new window a graph overview depicting the current selection region). IsaViz is a state of the art tool for browsing RDF

models.

Figure2.Newspaper schema in IsaViz Nevertheless,it has a rigid architecture which makes it dif-?cult to add application and/or scenario dependent opera-tions,i.e.other operations than the default ones supported by the tool.Figure2depicts the newspaper RDFS schema in IsaViz.

2.GViz

GViz[6]is a general purpose visual environment for browsing and editing graph-based data.Since RDF is es-sentially an attributed graph,one can use GViz to visu-alize RDF models.GViz’s chief advantage compared to most other graph visualisation tools is that it is easily cus-tomizable.In the past,GViz was customized with speci?c query and visualisation operations for reverse engineering data visualisation[6].Our experience was that this extensi-bility property with application-dependent operations is es-sential for producing effective data visualisations.Figure3 presents the newspaper RDFS schema in https://www.wendangku.net/doc/ca4849650.html,pared with Figure2,the IsaViz representation of the same model, the data structure is now easier to grasp.For the explanation of the used colors,see Sec.2.3.

In the following,a short description of GViz is given. 2.1.Data model

The data model we use is the RDF graph representa-tion.Nodes are RDF resources/literals and edges are

RDF

Figure3.Newspaper schema in GViz properties.The type attribute associated to a node speci-?es if a node is an AResource(anonymous resource),an NResource(a named resource is resource with a URI),or a Literal.Both nodes and edges have a value property that gives the associated RDF label.Note that the value for named resources and properties is a URI.The value of an anonymous resource has no related semantics.The value of literals is given by their associated string.As GViz’s data model is an arbitrary attributed graph,the above data are directly accommodated by the tool.

2.2.Operation model

GViz’s operation model comprises three main opera-tions:selection,graph editing,and mapping.Selection op-erations specify a set of nodes and edges from the original graph.Queries and?lters are thus naturally implemented as selections.Editing operations modify the graph data(struc-ture and/or attributes).Node/edge deletion or construction, graph metric computations,and graph layouts are thus im-plemented as editing operations that modify various parts of the data https://www.wendangku.net/doc/ca4849650.html,ing the observer pattern,all system components that depend on the changed data are automati-cally updated.The mapping operations map graph data to visual objects.Implementing different mapping operations corresponds to customizing the way the graphs are drawn. GViz’s architecture focuses on allowing users to easily de-?ne their speci?c operations.One such operation is the graph comparison,a useful feature if we consider e.g.an-alyzing the differences between two(RDF-based)mobile phone pro?les or the evolution of the pro?le schema.

2.3.Visualisation

In contrast to most graph visualisation tools we are aware of,GViz decouples the mapping from the graph layout op-eration.Graph layouts are attribute-editing operations that compute2D or3D positional attributes for nodes/edges. GViz uses different layouts among which we mention: spring embedder,directed(tree),3D stacked layout,and the nested layout.More information about the last two layouts can be found in[6].Furthermore,the visual appearance of nodes and edges in GViz is entirely https://www.wendangku.net/doc/ca4849650.html,ers can easily de?ne the shape,color,size,and other graphi-cal attributes of the node and edge‘icons’as function of their attributes.GViz’s approach to customization is to al-low users to provide callbacks,written in the Tcl scripting language,for most of its internal operations,mapping in-cluded.In all our scenarios,customizing the node or edge drawing amounted to writing an8to20lines Tcl callback that used the node and/or edge attributes to customize its drawing.

In the rest of this paper,all examples will be based on User Agent Pro?le(UAProf)[7],a CC/PP[2]vocabulary for describing mobile phone https://www.wendangku.net/doc/ca4849650.html,/PP vocabular-ies are RDFS representations for modeling device capabili-ties and user preferences.

Figure4presents the GViz graph representation of the UAProf schema.Graph nodes are depicted by rectangles and RDF graph edges are represented by fading lines.The lines are fading to the origin(subject)node so that the edge direction effect is created.We found that representing di-rectional information in this way is more effective than the classical arrow-drawing,as the latter produces too much vi-sual clutter for highly connected graphs.The node icons’colors convey the nodes’types:yellow for literals and green for resources.Three separate colors are used for edges:blue for edges with value rdf:type,red for edges with value rdfs:subClassOf,and white for edges with different value than rdf:type and rdfs:subClassOf.Note that,due to their loose coupling with other nodes,liter-als are positioned at the drawing’s periphery.The spring embedder layout naturally positions the most referenced nodes at the center:rdfs:Class and rdf:Property. As these nodes were selected with the mouse by the user, they are displayed in red by GViz instead of green.We also chose to represent nodes that have an edge with value rdf:Property with orange instead of green.As a consequence the only nodes that remained green are the Component node,its subclasses(describing the hardware and software platforms,the wap,push,and network charac-teristics,and the browser user agent),and rdf:Bag.Pro-ducing the above visualisation took about20minutes and amounted to writing three Tcl callbacks of less than40lines in total.

This visualisation allows one to easily distinguish the Component node and its subclasses,forming a“star with red rays”,and the rdf:Property and its instances,form-ing a“star with blue rays”.As a RDFS schema basically de?nes a set of properties to be used in the instance,a big cloud of orange nodes(property nodes)is present in the Fig-ure.Figure4enabled the users of our tool to see that the depicted UAProf schema(from10th of July2002)uses a wrong rdfs pre?x in rdfs:Property instead of rdf, a fact which was not discovered before this visualisation was

done.

Figure4.UAProf schema visualisation

3.Applications

We consider now four types of RDF-related applications: customizable selections

schema-instance comparison

instances comparison

schemas comparison

The last three applications are related to graph comparison. For graph(node value)comparison,we identify the spe-ci?c nodes(nodes only present in one of the models)and the common nodes(nodes present in all models).In com-paring graphs,it is important to distinguish between named resources and anonymous resources because the value of anonymous nodes has no semantics and should thus not be used in comparison.

3.1.Customizable selection

Figure5depicts the node representing the HardwarePlatform component,which was se-lected with the mouse by the user.The selection process is user-customized in the sense that the original edges that do not have the HardwarePlatform node as subject/object in Figure4are suppressed.This selection is similar to the radar view of IsaViz with the difference that it presents only the interesting edges(with respect to the selected node)instead of all edges from the original graph.As explained in Sec.2.3,the customization is done by letting the user specify the action GViz performs(in this case,the selection)by means of a Tcl script.Writing the script for our custom selection(of18lines of code)took less than5 minutes for a user familiar with GViz but not with

RDF.

Figure5.Selection in UAProf schema

3.2.Schema-instance comparison

A schema-instance comparison answers questions like: how much of the schema is instantiated in an instance?, what subpart of the schema is used by the instance?etc.To distinguish the resource types,we chose to represent named resources by triangle icons,literals by circles,and anony-mous resources(the resource standard)by rectangles.In contrast to the visualisation described in Sec.3.1,we now use color for comparison purposes,as described next.

Figure6shows the UAProf instance of a Nokia8310mo-bile phone.We use the grey color for anonymous nodes to stress that they are not to be compared.The nodes speci?c to the instance are yellow,the nodes speci?c to the UAProf schema are green,and the common nodes(i.e.present

in

Figure6.UAProf instance for Nokia8310 both schema and instance)are red.In Fig.6,we notice that most instance-speci?c nodes are the literals that character-ize this particular Nokia phone,such as e.g.Nokia8310, the phone name.Speci?c resources for the instance are rdf:Bag and the nodes that describe different components (the hardware and software platforms,the wap,push,and network characteristics,and the browser user agent).The common nodes(depicted in red)are the types of the com-ponents,since these appear in both instance and

schema.

Figure7.UAProf schema

Figure7describes the UAProf schema related to a Nokia

8310phone instance.As shown also in Fig.6,the common nodes(red triangles)are resources representing the compo-nent types.RDF is a semistructured language.An RDF instance doesn’t need to instantiate all properties of an as-sociated schema.As a consequence,we see the big cloud of green nodes which are schema speci?c nodes(nodes that are not appearing in the instance).

Finally,Figure8presents both the UAProf schema and the Nokia8310instance combined in one graph.This?g-ure is a combination of the previous two pictures.We no-tice that only a small part of the schema is instantiated by the instance(the common red part)and that this part con-sists of component types.Again,this type of insight in the RDF data was not attainable by the other RDF data brows-ing tools we

used.

Figure8.UAProf schema and instance for

Nokia8310

3.3.Instance comparison

Comparing several instances that validate the same schema answer questions like:what properties are speci?c in each instance?,what are the common properties of the instances?etc.Note that,by properties,we mean the value associated to a property.

Figure9compares the UAProf instances for four mobile phones:the(previous)Nokia8310,Ericsson T68,Erics-son T39,and Mitsubishi Trium.For this visualisation,we designed the following coloring scheme:instance-speci?c nodes are grey,the nodes shared by the two Ericsson phones are green,and the nodes common to all four phones are red. Looking at the four pictures we notice that their structure is roughly identical.This complies with the fact that they all instantiate the same schema.It is interesting to observe that all instances of a certain schema have the same structure which differentiates them from other instances.A useful ap-plication hereof is the visual identi?cation of instances that have the same(unknown)schema from an instance reposi-tory based on their structure.

We also noticed that there is only one common resource rdf:Bag,which immediately brings the question“where are the components?”We discovered that the reason for not having the components in the set of common nodes is that the Ericssons and the Mitsubishi use a previous version of the UAProf schema,which uses a different naming pre?x than the one used in the Nokia8310.Again,this fact was discovered only after the visualisation took place.Finally, the speci?c nodes are mostly represented by literals that characterize each mobile.Note that,being produced by the same company in the speci?c family of“T”mobile phones, the pro?les of the two Ericssons are very similar(large set of green

nodes).

Figure9.UAProf instances for four phones 3.4.Schema comparison

Comparing different versions of the same schema (schema evolution)enables one to better track the differ-ences among them.A visual representation of these dif-ferences answers questions like:which schemas are very similar to each other?which schema represents a major ar-chitectural break compared with the previous ones?etc.

Let us consider three UAProf schemas from2000,2001, and2002(the last one was already used in the previous sub-sections).Now we design the following coloring scheme: schema-speci?c nodes are grey;nodes in2000and2001 but not2002are green;nodes in2001and2002but not 2000are yellow;nodes present in all three schemas are red.

Figure10compares a UAProf version from2000with the UAProf version from2001.The large number of green nodes show that the2000and2001schemas have a lot in common,i.e.that the UAProf speci?cation is only mildly updated from2000to

2001.

Figure10.UAProf comparison:2000and2001

Figure11compares the UAProf version from2001with the UAProf version from2002.Nodes present only in2001 and2002(but not in2000)should appear in yellow.How-ever,a surprising discovery was that there were no yellow nodes.However,2002shows a lot of grey nodes(elements not present in2001,e.g.the push characteristics compo-nent,the Bluetooth pro?le).This means that the year2002 breaks the schema continuity present in2000and2001,i.e. it introduces many new elements.However,there are still overall similarities for the three years(the red nodes).A possible reasoning is e.g.that2002is the begin of a new product

family.

Figure11.UAProf comparison:2001and20024.Conclusions

In this paper,we discussed the usage of GViz,a general purpose graph visualisation platform,for the RDF graph vi-sual https://www.wendangku.net/doc/ca4849650.html,pared to other RDF data browsing tools,we were able to produce visualisations that answered more complex questions about the data and give a more ef-fective insight in the data structure.The produced visualisa-tions easily answered queries such as:which schema parts are present in an instance,which properties are speci?c to a given instance in an instance set,and how do schemas evolve in time.An interesting result was the discovery of (unexpected)facts about the examined data,which were simply not apparent during browsing with other RDF tools.

From an application design point of view,our experi-ence with GViz was very positive.The tool’s mechanism of providing customization of its selection,visualisation, and query operations by user-written Tcl callback scripts proved highly versatile and allowed us to program and?ne-tune new visualisation scenarios in minutes.This fact is worth mentioning,as few tools(for graph visualisation in general,and for RDF data in particular)provide such?exi-bility,which we deem to be essential for adapting a general-purpose tool to a speci?c scenario.This lack of?exibility may be one of the main(though not often discussed)rea-sons for which we see much less reuse of relational data visualisation tools as compared to e.g.the more classical scienti?c data visualisation tools.

We next plan to look at more RDF data exploration ap-plications,such as RDF graph editing operations(GViz of-fers support for graph editing).We also plan to investigate useful metrics and?lters to be applied for an RDF graph. References

[1]S.Card,J.Mackinlay,and B.Shneiderman.Readings in In-

formation Visualization.M.Kaufmann,1999.

[2]G.Klyne,F.Reynolds,C.Woodrow,and https://www.wendangku.net/doc/ca4849650.html,-

posite capability/preference pro?les(cc/pp):Structure and vocabularies.W3C Working Draft08November2002. [3]S.C.North and E.Koutso?os.DOT and NEATO user’s

guide.AT&T Bell Labs Reports,1996.http://www.

https://www.wendangku.net/doc/ca4849650.html,.

[4]N.F.Noy,M.Sintek,S.Decker,M.Crubezy,R.W.Ferger-

son,and M.A.Musen.Creating semantic web contents with protege-2000.IEEE Intelligent Systems,16(2):60–71,2001.

[5] E.Pietriga.Isaviz:a visual environment for browsing and

authoring rdf models.The Eleventh International World Wide Web Conference(Developer’s day),2002.

[6] A.Telea,A.Maccari,and C.Riva.An open toolkit for proto-

typing reverse engineering visualization.In IEEE EG VisSym ’02,pages241–250.Eurographics,2002.

[7]Wireless Application Protocol Forum,Ltd.Wireless applici-

ation group:User agent pro?le.Version20October2001.