The Web of Data has grown to a size of several billion triples and provides human interface opportunities and challenges beyond those of the traditional Web. Although Linked Data is now generated at a fast pace and very large scale, we observe that browsing and visualisation of Linked Data is still in its infancy. In parallel to the enormous boost in Linked Data, recent work on integrating 3D graphics capabilities into the W3C technology stack provides fresh momentum for an e ort to extend the Web with 3D content and technologies. In light of the timid uptake and consumption of Linked Data by non-technical audiences, we make the case for Web3D-based user interfaces to the Web of Data and aim at promoting synergistic research in the Web3D and Web of Data communities. To that end, we describe a scenario that requires the combination of Linked Data from geospatial and encyclopedic data sources and the transformation of the combined Linked Data into a format amenable to Web3D rendering. Based on the scenario, we derive highlevel requirements and propose a Linked Data-driven design pattern based on REST architecture principles that satis es the requirements. Our prototypical implementation shows that a combination of current Web technologies is su cient to implement distributed application that ultimately arrive at Web3D renderings of Linked Data. Based on the experiments, we identify and discuss areas which require further research.
The World Wide Web has irreversibly changed the way we collect, validate and disseminate data and information. The Web is built on the idea of hyperlinking documents into a single global document space, hereby providing the fundamental principles for accessing, navigating and publishing Web documents. Until recently, however, the principles that promoted the Web of Documents have not been applied to the abundance of data related to media, bibliography, encyclopedical knowledge, or geo sciences. Historically, these data sets have been made available as documents in various Copyright c 2012 for the individual papers by the papers’ authors. Copying permitted only for private and academic purposes. This volume is published and copyrighted by its editors. Dec3D workshop at WWW2012, Lyon, France content formats or from databases accessible through Web APIs. Lacking uniform identi cation, retrieval and representation mechanisms, Web data sets have been fragmented into disconnected data silos.
To publish, access and integrate data on the Web, BernersLee [4] devised the Linked Data Principles, a set of best practices based on a small set of standards, i.e., Uniform Resource Identi ers (URIs), the Hypertext Transfer Protocol (HTTP), the Resource Description Framework (RDF) as well as a number of serialisation formats. By establishing typed data-level links between items from disparate data sources, Linked Data facilitates the creation of a single global data space, enables the integration of data from disparate sources and ultimately provides expressive query capabilities.
Automated production and interweaving of Linked Data at a large scale has now grown the Web of Data to a size of 25.2 billion triples1 and an increasing number of applications demonstrates the potential of exploiting the Web of Data. For example, the GADM-RDF2 project contains the world's administrative boundaries and interlinks to existing spatial datasets. The DBpedia3 project [2] constantly derives a Linked Data corpus from the Wikipedia encyclopedia and describes more than 3.64 million interlinked entities, containing factual data on persons, places and organisations. The growing Web of Data, however, provides human interface opportunities and challenges beyond those of the traditional Web. A signi cant challenge is found in the formalised RDF encoding of Linked Data. RDF, a format created for machine consumption, limits the use of Linked Data to those capable of reading and interpreting raw RDF. Even for a technical audience, the lack of visual aid impedes the identi cation of relevant information and the unearthing of previously unseen correlations. Aiming at improved accessibility and usability of Linked Data, a limited number of Linked Data browsers and visual search engines o er both text-based and visual presentation options. The Tabulator browser [5], for example, provides several conventional user interface metaphors such as a timeline and a tabular view. VisiNav[?], a system based on an interaction model designed to search and navigate large amounts of Linked Data, o ers amongst other things a 2D graph visualisation 1http://www4.wiwiss.fu-berlin.de/lodcloud/ 2http://gadm.geovocab.org/ 3http://dbpedia.org/ of Linked Data items. Although Linked Data is now generated at very large scale, we observe that visualisation and browsing of Linked Data is still in its infancy.
In parallel to the enormous boost in Linked Data
technologies, recent work [
Our contribution is in showing how to combine Linked Data and Web3D functionality accessible via RESTful interfaces into a coherent system which allows for the querying, combination and transformation of Linked Data into a format amenable to Web3D rendering.
The remainder of the paper is organised as follows. In Section 2 we cover related work, and in Section 3 we derive and describe several high-level requirements a Linked Datadriven Web3D application should ful l. We provide an application prototype as well as a description of its essential components in Section 4. Furthermore, we discuss the interplay of data-providing and data-processing RESTful services. In Section 5 we discuss our lessons learned, point out limitations and suggest areas which require future work. We conclude with summary in Section 6.
There are several approaches in the literature that describe the design and functionality of user interfaces for the Web of Data. These user interfaces can be grouped into systems that solely provide text-based functionality and those that include visual presentation options.
We start by reviewing systems that use textual representations of Linked Data items and their relationships. Dipper4 provides a Linked Data endpoint for retrieval and browsing of Linked Data from a set of Linked Data repositories. Where available human-readable RDF labels are used to format the results. The Disco Hyperdata Browser5 allows the user to navigate the Web of Data and renders all information as a HTML table of property-value pairs. Where available, human-readable labels are used to format the results. As for Disco, Marbles6 allows navigating the Web of Data. Marbles strives to improve the user experience by formatting Linked Data for presentation to end users; di erent colours are used to distinguish the sources of the information retrieved. Piggy Bank7, a Firefox extension, allows the extraction of Linked Data from web sites and enables users to perform faceted search and browsing across Linked Data collected by multiple users. Again, results are presented using list and tabular interfaces.
User interfaces o ering visual presentations are provided by more advanced RDF browsers or specialised Linked Data applications. For example, the Tabulator browser [5] provides several conventional user interface metaphors such as a timeline and a map view. RDF Gavity8 is a tool for visualising ontologies on top of the JUNG Graph API and supports user-speci ed graph lters for speci c views. Browsers such as Fen re9 or FOAFNaut10 visualize the relationships among Linked Data items employing a graph visualisations. VisiNav[?], a system designed to search and navigate large amounts of Linked Data based on an interaction model, offers amongst other things a 2D graph visualisation of Linked Data items and their relationships.
Although Linked Data is now generated at very large scale, we observe that visualisation and browsing of Linked Data is still in its infancy. With the exception of DBpedia Mobile11, a location-centric DBpedia client application for mobile devices, most of the reviewed systems focus on supporting navigation and exploratory discovery in the overall Web of Data. Visualisation techniques for a detailed inspection and analysis of Linked Data query results, i.e., a user's particular region of interest in the Web of Data, are typically restricted to tabular or list presentations. In contrast to these existing systems, we provide a proof-of-concept prototype illustrating how to achieve Web3D visualisations of Linked Data query results in a completely distributed architecture.
Following Hausenblas [12], a Linked Data-driven Web application consumes and potentially manipulates Linked Data sets and o ers appropriate interfaces and content for both 4http://api.talis.com/stores/iand-dev1/items/ dipper.html 5http://www4.wiwiss.fu-berlin.de/bizer/ng4j/disco/ 6http://www5.wiwiss.fu-berlin.de/marbles 7http://simile.mit.edu/wiki/Piggy_Bank 8http://semweb.salzburgresearch.at/apps/ rdf-gravity/ 9http://fenfire.org/ 10http://www.foafnaut.org/ 11http://wiki.dbpedia.org/DBpediaMobile humans and machines. We extend this de nition towards Linked Data-driven Web3D applications in a straight-forward way and discuss several high-level requirements for Linked Data-driven Web3D applications in the following. Unifying Web3D and Linked Data representations: The description and augmentation of 3D models and virtual spaces with semantic metadata and ontologies has been in the focus of the 3D modeling community for a long time, yet, the issue is far from resolved. The Linked Data community, however, successfully facilitates this task for disparate data sources by establishing typed data-level links between data items. We believe that several strategies exists in order to, at least partially, transfer the Linked Data principles to the Web3D. The conventional approach, cf. Pittarello et al. [14], is to describe the semantics of Web3D content in annotation tags provided by the Web3D speci cation or by means of RDFa12 annotations at the tag attribute level. In order to bene t from such semantic annotations, the initial extraction and subsequent management of semantic scene annotations is required. A very radical concept following Kalogerakis et al.[13] is to accept RDF as the single data model for publishing 3D content on the Web and to merge graphics content and scene semantics into the Web of Data. However, this leaves open pressing questions on how to e ciently render and manipulate graphics content in its RDF encoding. As a third stance and as a trait of Linked Data-driven Web3D applications, we adopt REST principles throughout and retrieve Web3D content as REST representations of Linked Data resources by content-negotiation. Although an established REST design practice, we feel that this concept requires further thought.
On-demand integration: The diversity and size of Linked Data sets make it di cult { even in our prototype application { to just fetch and integrate Linked Data and relevant geometries from all sources and load the combined Web3D scene graph into a Web browser for rendering. Given that source datasets are often distributed, we require the overall Linked Data-driven Web3D application (dispersed across multiple providers of data, functionality and client-side rendering) to access and integrate data on demand using the pull model. Existing systems that demonstrate initial results for integrating semantics into virtual environments suffer from scalability and real-time performance issues. Our aim is to use caching and asynchronous communication to achieve performance conductive to interactive visualisations; our initial experiments indicate that our system requires more optimisations to achieve the goal.
Integration of content and functionality: Accepting the preceding argument of on-demand integration, a Linked Data-driven Web3D application needs to integrate both content and functionality. Examples for content are the GADMRDF or DBpedia datasets; examples for functionality are 3D meshing and shading services. The various components have to be connected together in a way that achieve the intended application purpose. We see the Web as a platform and want to combine services that are decentralised. 12http://www.w3.org/TR/xhtml-rdfa-primer/ Declarative integration: We aim to use declarative means to describe data and links between the data, to arrive at an ecosystem in which data providers interlink their datasets without the need for direct coordination. An ideal system would gure out the data sources to combine with dedicated data processing services to ultimately arrive at the stated goal of a user. However, such a system has proven elusive. We thus opt for a declarative coordination of the interoperation between data sources and functionality. Interactive visualisations: Typically, Linked Data-driven Web3D applications require that user input is possible, e.g. for selecting a visual encoding. For an interactive application, the overall system must provide high-throughput and high update rates with low latency. While these qualities require further optimisations of the underlying Linked Data and Web3D technologies, the overall user-perceived latency can be improved using techniques such as incremental content loading or user reassurance during wait times.
In the following, we describe a straightforward Linked Datadriven Web3D application leveraging the emerging 3D graphics capabilities of the W3C technology stack as well as the increasing amount of Linked Data available on the Web. Given statistical data from DBpedia (see Figure 1(a)) and geospatial data from GADM-RDF (see Figure 1(b)), we aim at generating an interactive 3D choropleth map [9] visualising the world distribution of wealth (see Figure 1(c)). We call our prototype application \Hello (3D) World!" and believe that its interactive Web3D interface allows the user to e ectively grasp the meaning of the underlying Linked Data. Realising the application involves several steps: querying across data sources; transforming query results into a Web3D scene graph; and rendering the Web3D scene graph. Based on REST principles, our prototype declaratively combines Linked Data into a format amenable to 3D rendering. To this end, we use RESTful services13 for basic computer graphics, e.g. shading, meshing and coordinate transformations, SPARQL14 for queries spanning multiple disparate Linked Data sources containing heterogeneous data, and XQuery15 for executing the overall application logic.
Antedating a more detailed description of all involved artifacts, Figure 2 demonstrates our Linked Data-driven design and depicts the following interaction patterns: 1. Via GADM-RDF 's Linked Data Services[16], resolve Linked Data resources of the world's nations and retrieve RDF including (i) URIs for their KML geometries and (ii) owl:sameAs'16 to interlinked data sets. This corresponds to executing the SPARQL query (cf. Listing 2) against GADM-RDF. Cache the results locally. 2. Resolve each ?geo URI from Step 1, parse the retrieved
KML data and send each < kml : P olygon > to 13http://jersey.java.net/ 14http://www.w3.org/TR/rdf-sparql-query/ 15http://www.w3.org/TR/xquery/ 16owl:sameAs links indicate that two URI references actually refer to the same thing. (a) DBpedia on the world's GDP. (b) Geo data in GADM-RDF. (c) Hello (3D) World! the REST service Geo2Ecef transforming WGS84 into ECEF coordinates. Post the transformed polygons to the REST service Poly2Mesh which in turn responds with fresh URIs for each generated XML3D mesh. The results are cached locally. 3. Using the owl:sameAs's from Step 1, execute the query shown in Listing 3 against DBpedia and retrieve each nation's gross domestic product based on the purchasing power parity. Post the ?gdp literals to the REST service Gdp2Shader which responds with fresh URIs for XML3D shaders. Cache the results locally. 4. Use the URIs generated in Step 2 and Step 3 and generate an XML3D-compliant scene graph. 5. Return the generated XML3D-compliant scene graph to the browser for rendering and display.
In the following, we describe the general high-level components of Linked Data-driven Web3D applications using the example of \Hello (3D) World!".
Linked Data-driven Web3D applications require some runtime environment for implementing their business logic. The business logic coordinates the invocation of the various data sources and services in the right order to transform and adapt the content into a 3D scene according to the specications of the user. Such runtime environments should enable access to various distributed data sources using various languages, permit highly granular selection based on both the hierarchy and ordering of structured data, and allow for complex data transformations that may involve hierarchical restructuring.
Developed by the XML Query working group of the W3C,
XQuery is de ned as a query language for XML data. Path
expressions can be used to select based on XML structure
(both hierarchical and sequential) and XML result
structures can be easily generated. The XQuery Update Facility
provides the ability to update data or to add annotations
and metadata; specialised capabilities beyond the large
selection of built-in functions and operators are provided via
an extension mechanism through function libraries. XQuery
supports library modules, i.e., collections of function de
nitions and global variable declarations. The XQuery engine
imports a module once during compilation, the compiled
module is then made available through the static XQuery
context. In particular, we make use of the EXPath
httpclient module17. This module allows one to deal with most
aspects of the HTTP protocol, and hereby enables the
invocation of RESTful data providing and processing services.
XQuery appears especially alluring given the fact that
XSPARQL18, a merge of SPARQL and XQuery providing
concise solutions for mapping between XML and RDF in
either direction, is de ned as an extension of XQuery's
formal semantics adding a few normalization mapping rules.
In essence, XSPARQL queries are reduced to XQuery with
interleaved calls to a SPARQL engine via the SPARQL
protocol [
Linked Data-driven Web3D applications consume and manipulate various data sets and o er appropriate interfaces and content for both humans and machines. While we strive to ultimately arrive at 3D renderings of complex data, some technology for the retrieval, combination and manipulation of data stored in disparate sources is needed. The query language SPARQL can be used to express queries across diverse Linked Data sources. Its syntax resembles SQL, but SPARQL is far more powerful, enabling queries spanning multiple disparate data sources containing heterogeneous data.
Our prototypical implementation employs SPARQL in order to retrieve Linked Data of the world's nations from GADM17http://www.expath.org/modules/http-client/ 18http://www.w3.org/Submission/ xsparql-language-specification/ 19http://code.google.com/p/rdfquery/ RDF . The corresponding query shown in Listing 2. In addition, each nation's gross domestic product based on the purchasing power parity is retrieved from DBpedia using the SPARQL query shown in Listing 3. 1 PREFIX s p a t i a l : <http : / / geovocab . org / s p a t i a l#> 2 PREFIX ngeo : <http : / / geovocab . org / geometry#> 3 PREFIX owl : <http : / /www. w3 . org /2002/07/ owl#> 4 SELECT ? n a t i o n ? geo ? l i n k 5 WHERE f 6 ? n a t i o n rdf : type s p a t i a l : Country . 7 ? n a t i o n ngeo : geometry ? geo . 8 ? n a t i o n owl : sameAs ? l i n k 9 g
Listing 2: SPARQL query against GADM-RDF 1 PREFIX dbpedia2 : <http : / / dbpedia . org / p r o p e r t y/> 2 SELECT ? gdp 3 WHERE f 4 ? l i n k dbpedia2 : gdpPppPerCapita ? gdp . 5 g
geo2ecef. This service is implemented to accept a kml:Polygon, to convert its geodetic coordinates into Earth-centred earthxed coordinates, and to respond with results in XML format. In the following, we give an exemplary GET request and the service's response.
GET /geo2ecef?input=
'<kml:Polygon>x1,y1,z1...</kml:Polygon>' HTTP/1.1 Host: http://www.host.org Accept: application/xml HTTP/1.1 200 OK Server: http://www.host.org Content-Type: application/xml <ecefPolygon>f(x1,y1,z1)...</ecefPolygon> gdp2shader. This service accepts numerical values for the generation of xml3d:shader elements to be resolved by a Linked Data-driven application. The xml3d:shader element is derived from the service input and a prede ned colour scale. Listing 4 shows XML3D code generated by this service. Anticipating more complex shader generation, the service is implemented as POST and responds with Linked Data containing fresh URIs for xml3d:shader elements. Additionally, this service provides a GET method for shader retrieval.
Our prototype invokes Web3D REST services performing 21 <<sf hloaadte3r nsacmr ie=pt"=d"i uffruns:exCmol3l do r: "sh>a0d.0e8r : p0 h.0ong0."0></ f l o a t 3 > preprocessing and mapping operations. Invocation of the 3 <f l o a t name=" a m b i e n t I n t e n s i t y ">0.4</ f l o a t > preprocessing service geo2ecef performs coordinate trans- 4 </shader> formations, whereas the mapping services poly2mesh and gdp2shader produce xml3d elements. All Web3D services Listing 4: Generated XML3D shader element are implemented using JAX-RS and Jersey and deployed to Apache Tomcat Servlet containers. Adhering to Fielding's architectural principles [7], we designed these services in the following ways. poly2mesh. When receiving POST requests, this mapping service accepts kml:Polygons, spawns triangulation threads for POSTed kml:Polygons and immediately returns Linked Data containing fresh URIs. These URIs identify xml3d:mesh elements being computed using a Delaunay triangulation backend20 and can be resolved by Linked Data-driven applications. The service additionally provides a GET method for xml3d:mesh retrieval; an exemplary xml3d:mesh element generated by this service is shown in Listing 5. 1 <mesh type=" t r i a n g l e s "> 2 <int name="index ">0 1 2 1 3 2</int> 3 <f l o a t 3 name=" p o s i t i o n " >0.96 . . . 0.13</ f l o a t 3 > 4 <f l o a t 3 name="normal ">0.0 . . . 1.0 </ f l o a t 3 > 5 <f l o a t 2 name="texcoord ">1.0 . . . 0.0 </ f l o a t 2 > 6 </mesh>
Being focused on the task of conveying complex Linked Data to the user, we aim at rendering and displaying 3D visual20http://code.google.com/p/poly2tri/ isations in the browser. Via the browser, users have the ability to change various parameters that the application designer has provided. The input from the user interface triggers queries to the execution engine which in turn combines content from Linked Data sources and functionality from services into a Web3D scene graph. The Web3D scene graph is handed to the rendering and displaying environment for generating the visual application output. We decided for XML3D as rendering and displaying environment for its clean and comprehensible speci cation; although we sadly miss support for external references in the current implementation. As future work, we would like to assess alternative Web3D proposals with respect to ease of use, performance and scalability.
We now discuss lessons learned based on the experiences gained during experimentation and put our requirements into perspective.
The success of implementing a Linked Data-driven 3D Web Application crucially depends on the availability of highquality data. In our scenario, we depend on a complete set of links between countries in GADM-RDF and DBpedia. While GADM initially provided some links, both datasets were only sparsely interlinked. Only after completing the linkage we were able to query both datasets in combination. The size of the geometries in GADM-RDF were another stumbling block. Geometries of countries can reach several MB, and merely transmitting these les takes time. We solved the issue by accessing more coarse-grained versions of the geometries. For the triangulation, which is a relatively time-consuming calculation, we opted for a service which queues the required computation steps and asynchronously processes the input. We use caching throughout (via HTTP's Last-Modi ed header and a Squid cache21) to avoid costly re-computation.
We use SPARQL queries and XQuery programs to access data and coordinate computation, which allows for a concise speci cation of the various computation steps. We used SPARQL to query the data in combination, and XQuery to invoke functionality, as XQuery engines can be extended with user-de ned functions. However, using two separate query engines is suboptimal; an execution environment which natively supports RDF queries and XQuery-like programming constructs would be our more elegant system of choice. Ideally, such an interpreter of data integration programs would run client-side in a web browser, to remove the bottleneck which a centralised query processor represents. Whether XSPARQLis a suitable candidate for such a language is subject to future work.
Finally, the lack of maturity of the deployed Web3D implementation required some workarounds; for example, the XML3D speci cation allows for providing URIs which the browser would dereference and include in the scene graph. Such functionality would t our distributed architecture very well, however, current implementation lacks these network lookups, and thus we need to construct a large monolithic le with the scene graph rather than having a scene graph le which can include network-accessible references.
The overall goal of Linked Data-driven Web3D applications is to enable the end user to query, understand and interact with often complex and incomprehensible Linked Data in more user-friendly and e cient ways. Suitable and coherent Web3D visualisations may help in conveying complex Linked Data, support users in sense making and information exploration and discovery, and eventually provide descriptive, intuitive and e ective user interfaces to the Web of Data. To facilitate the speci cation and implementation of such Linked Data-driven Web3D applications, we provide an architecture blueprint that enables the integration of distributed content and functionality into coherent Linked Data-driven applications with Web3D user interfaces.
[