1. Journal Club Report “Linked Data – The Story So Far” Denise Warzel Feb. 2012

2. Linked Data – The Story So Far Authors: Christian Bizer, Frele Universität Berlin, Germany Tom Heath, Talis Information Ltd, United Kingdom Tim Berner-Lee, Massachusetts Institute of Technology, USA International Journal on Semantic Web and Information Systems (2009) Volume: 5, Issue: 3, Publisher: Elsevier, Pages: 1-22, DOI: 10.4018/jswis.2009081901 10.4018/jswis.2009081901 Follow on to Berners-Lee talk: http://www.ted.com/talks/tim_berners_lee_ on_the_next_web.html These authors are thought leaders for Linked Data Good overview of the state of the art in Linked Data Linked Data uses the technologies of the Semantic Web Describes the various technologies and tools caBIG has not really taken advantage of the Semantic Web Hope to stimulate discussion and consideration of this approach for publishing data

3. Paper Overview Section 1: Introduction Section 2: Overview of key features of Linked Data Section 3: Activities and outputs of the Linking Open Data Project Section 4: State of the art in publishing Linked Data Section 5: Overview of the Linked Data applications Section 6: Compares Linked Data to other technologies for publishing structured data on the Web Section 7: Ongoing research challenges

4. 1. Introduction “Linked Data refers to a set of best practices for publishing and connecting structured data on the Web” Similar to principles for publishing web documents Web documents are linked via hypertext links Web browsers allow navigation from one document to another Search engines index the documents and infer potential relevance to users' search queries These kinds of links do not describe the nature of the link between two documents, HTML is not expressive enough, the relationship is implicit Traditionally data on the web is made available as raw dumps CSV XML HTML Tables 4

5. 1. Introduction Web has evolved to include links to documents and data Linked Data principles are for publishing data on the web New way for sharing data Publishing and connecting structured data Creates a “web of data” Mechanism similar to hypertext links in web documents Connecting data from diverse domains such as people, companies, books, scientific publications, films, music, television and radio programes, genes, proteins, drugs and clinical trials, online communities, statistical and scientific data Generic Linked Data Browsers and Search engines will allow users to crawl the web of data by following links between data sources start browsing in one data source and navigate links to other data sources

6. Linked Data differs from Web 2.0 Mashups Mashups rely on fixed data sources put together to present new or related information Portal Pre- processing Web 2.0

7. Linked Data differs from Web 2.0 Mashups In Linked Data new information automatically becomes available as users deliver new data sources to the web that are linked to existing data sources Users will start in one data source and by following links may end up in an entirely different data source Heavier lines indicate more links between the two datasets Much of the linked open data cloud was generated by putting wrappers around existing databases Some data comes from info boxes seen on right hand side of Wikipedia articles Linked Open Data Cloud

8. Linked Data relies on documents containing data in Resource Description Framework (RDF) format making it machine-readable You can build links between data from different sources in different geographical locations from different organizations Using this format you can make typed statements about the links between things “Typed” meaning that certain RDF tags mean certain things, there is explicit meaning conveyed that RDF query languages can interpret RDF provides a generic, graph-based data model with which to structure and link data on the web RDF Triples consist of 3 parts: Subject - Predicate – Object The Predicate specifies how the Subject and Object are linked Explicit linkage using explicit language 2. What is Linked Data?

9. Linked Data Principles “Rules” for publishing linked data (Berners-Lee 2006) 1.Use Uniform Resource Identifiers (URIs) as “names” for things. (all three parts of the RDF Triple are represented by a URI) 2.Use Hypertext Transfer Protocol HTTP so that people can look up the URIs (“names”) 3.When someone looks up a URI provide useful information using the standards (RDF, SPARQL) 4.Include links to other URIs so that they can discover more things 9

10. RDF Triples Person A Knows Person B subjectobject predicate Predicates specifies how the subject and object are related. URI Using URIs, A, B and the Predicate can be in different data sets on the web.

11. Linked Data – another example Person C Author Scientific Article D subject object predicate URI Person A Knows Person B subjectobject predicate URI “RDF Vocabulary”

12. Linked Data Inferencing allows implicit relationships to be discovered URI subject object predicate URI subjectobject predicate Figure derived from text in the article Blue text added for emphasis

13. “RDF Triples” http://www.rdfabout.com/quickintro.xpd @prefix :. :john a :Person. :john :hasMother :susan. :john :hasFather :richard. :richard :hasBrother :luke. RDF/XML representation of the above: “RDF Vocabulary”

14. The Linked Data Technology Stack Three key techologies: URIs, HTTP Protocol, RDF URIs identify resources (subject, predicate, object) HTTP protocol retrieves information about resources RDF statements describe resources Builds on the classic document Web architecture Linked Data can contain any type of data Anyone can publish data Data publishers are unconstrained in choice of RDF vocabularies (the terms used for predicates) URIs are connected by RDF links, creating a global data graph that spans data sources thus enabling the discovery of new data sources

15. The Linked Data Technology Stack From application development perspective the Web of Data has the following characteristics: Data is strictly separated from formatting and presentational aspects Data is self-describing If an application consuming Linked Data encounters data described with an unfamiliar RDF vocabulary (predicates), the application can deference the URIs that identify the vocabulary terms in order to find their meaning. The use of HTTP as a standard data access mechanism and RDF as a standard data model (subject/predicate/object) simplifies access compared to web APIs which rely on heterogeneous data models and access interfaces Data is Open Unlike mashups, applications do not have to be implemented against a fixed set of data sources, but can discover new data sources at run-time by following RDF links

16. About dereferencing URIs …. - the act of retrieving information about a resource identified by a URI When information about the Subject URI “…/data#DIG” is dereferenced (acted upon) the dig.csail.mit.edu server answers with a RDF description about the resource identified by the “…/data#DIG”  the server provides a set of RDF statements that say it is “MIT Decentralized Information Group” (as opposed to delivering an HTML page) When the Object URI is dereferenced “…Berners-Lee/card#1” the W3C server provides a set of RDF statements describing the resource identified by the “…Berners-Lee/card#1” When the Predicate URI is dereferenced “…foaf/member” the XMLNS server returns RDF statements providing a definition of “…foaf/0.1/member”.

17. The RDF Triple This statement connects the subject “… film/77” in the Linked Movie Database (data.linkedmdb.org) with the description provided by Dbpedia.org, The predicate states that the URI “…film/77” and the URI “…Pulp_Fiction_%20film%29” refer to ’”the same thing”. This Predicate is represented by the URI “….owl#sameAs” which is a Web Ontology Language (OWL) term. OWL is used as the RDF vocabulary (predicate) in this RDF statement, just as FOAF was basis of the RDF Vocabulary term “member” in the prior slide.RDF statement.

18. 3. The Linking Open Data Project http://www.w3.org/wiki/SweoIG/TaskForces/CommunityProjects/LinkingOpenData - Bootstrap the “Web of Data” [cloud] by identifying existing datasets that are available under open licenses, converting them to RDF according to Linked Data principles and publishing them on the web - Initial involvement by BBC, Thomson Reuters and the Library of Congress - Anyone can participate by following the Linked Data principles - DBPedia (wikipedia info boxes) and Geonames (RDF about geographical locations around the world) are two key hubs in this web of data that contains content about : http://www.w3.org/wiki/SweoIG/TaskForces/CommunityProjects/LinkingOpenData Geographic locations People Companies Books Scientific Publications (PubMed) Films Music Television and Radio programmes Genes (GeneID, GeneOntology) Proteins (UniProt) Drugs Clinical Trials Online communities Statistical data Census results Reviews

19. 4. Publishing Linked Data on the Web Three Basic Steps to publishing data that becomes part of the “Linked Data Cloud”: 1.Assign HTTP URIs to the entities to be described in the data set and provide servers for dereferencing these URIs into RDF statements 2.Make RDF statements (links) about other data sources on the web so clients can navigate the Web of Data as a whole by following the links 3.Provide metadata about published data so clients can assess the quality of published data (dataset metadata)

20. Choosing URIs URIs are used for the subject, predicate and object (bears repeating) Different parties may publish information about the same real-world entity using different URIs – URIs are “aliases” for the real-world entity E.g. Berlin http://dbpedia.org/resource/Berlin http://sws.geonames.org/2950159 URI aliases allow different information providers to speak about the same entity Information providers set owl:sameAs links to URI aliases they know about Publishers must choose the URI schema for the Subject and Object, and an RDF vocabulary for the predicates

21. Choosing URIs, how does it work? By convention, the use of two HTTP URI patterns for linked data help client software distinguish between URIs for real-world entities (a document) versus URIs that are used for web documents (RDF) 303 URIs – http://www.example.com/id/alice “303 See Other” - The 303 is a redirected to the real URI - through content negotiation selects the RDF document Hash URIs - http://www.example.com/about#alice Used for resources that are not HTML documents Clients know to strip off the fragment represented at the hash before retrieving the URI Through content negotiation information in the HTTP Header specifies what kind of media/formats the user can accept, allows the selects the RDF document Content Negotiation

22. Choosing RDF Vocabularies for Linked Data Different communities have specific preferences for publishing data on the web Good practice to reuse vocabularies from well-known RDF vocabularies FOAF (Friend of a Friend) SIOC (Semantically-Interlinked Online Communities) SKOS (Simple Knowledge Organization System) DOAP (Description of a Project) vCard - Electronic Business cards, has become an ontology Dublin Core OAI-ORE -Open Archives Initiative Object Reuse and Exchange GoodRelations – Professional Web ontology for E-Commerce If new terminology is developed, it should be self-describing using URIs to identify the terms and be dereferencable Common serialization format for linked data is RDF/XML, or Notation3 and Turtle if human inspection of RDF data is required These practices makes it easier to allow clients to retrieve and process linked data

23. Link Generation Build data source including RDF links (statements) to point to other URIs in other data sources that are part of the Web of Data Choose a URI naming schema for the Subject and Object Publications - ISBN and ISSN numbers Finance – ISIN identifiers Products - EAN and EPC codes Life Sciences – identification schemata for genes, molecules and chemical substances (HUGO, UNIPROT, GENEID) Use automated or semi-automated approaches to generating RDF statements This approach was used to generate links between data sources in the LOD cloud When link source and target data sets already both support one of these schema, implicit relationships between entities can be made explicitly by navigating the RDF links (same RDF Vocabulary)

24. Link Generation When no shared schema exists, RDF statements can be generated based on the similarity of entities in both data sets Examples of mechanisms to generate links Similarity computations (Elmagarmid et al., 2007; Raimond et al., 2008) Such as in music comparing the names of artists as well as the titles of albums and songs Duplicate detection (Winkler, 2006) Ontology matching (Euzenat & Shvaiko, 2007) RDF link generation frameworks are available Silk Framework lets data publishers set links between their data source to other data sources LinQL Framework is an extension of SQL that integrates querying with link discovery methods

25. Metadata Several types of metadata increase utility for data consumers Creator, creation data, creation method Provenance information using Dublin Core terms of Semantic Web Publishing vocabulary Open Provenance Model provides terms for describing transformation workflows (wasGeneratedAt, wasGeneratedBy, wasEncodedBy, etc) Methods are being designed to provide evidence (maybe expressed as RDF statements) for how RDF links change over time Technical metadata Information about the data set and its interlinkage relationships with other data sets Vocabulary of Interlinked Datasets (VoID) defines terms and best practices to categorize and provide statistical meta-information about datatsets What about Metadata about the data?

26. Publishing Tools Serve RDF Content or Provide Linked Data views over non-RDF legacy data sources Views shield publishers from dealing with technical details Supports dereferencing URIs into RDF descriptions D2R Server Publisher defines a mapping between relational db schema and target RDF vocabulary D2R publishes a Linked Data view over the db that supports SPARQL queries Virtuoso Universal Server Serves RDF data via a Linked Data interface and a SPARQL endpoint Stores RDF directly or created on the fly from non-RDF relational db based on publisher supplied mapping Talis Platform Software as a Service (SaaS) accessed over HTTP Native RDF storage accessible as a SPARQL endpoint and REST APIs

27. Publishing Tools Pubby An extension to any RDF store that supports SPARQL Rewrites URI requests into SPARQL DESCRIBE queries against RDF store [defererences URIs] Also provides HTML view over the datastore and handles the 303 URI redirects and content negotiation [serving the format that the client/user agent can accept e.g. PNG vs GIF to a web browser that only accepts one type of format] Triplify Extends existing Web applications with Linked Data front end Based on SQL query templates, Triplify serves a Linked Data and JSON view over the database SparqPlug Enables the extraction of Linked Data from legacy HTML documents Serializing the HTML as RDF and allows users to define SPARQL queries that transform elements into an RDF graph of their choice

28. Publishing Tools OAI2LOD Server Linked Data Wrapper for document servers that support Open Archives OAI- RMH protocol SIOC Exporters (SIOC - Semantically Inter-linked On-line Communities) Linked Data wrappers for several blogging engines, content management systems and discussion forums such as WordPress, Drupal and phpBB

29. 5. Linked Data Applications Applications that exploit the Web of Data Three categories: Linked Data browsers Linked Data search engines and indexes Domain specific Linked Data applications

30. Linked Data Browsers Linked Data Browsers – Disco, Tabulator, Marbles, FOAFNaut and Fenfire Navigate between data sources by following links expressed as RDF triples Eg. View Dbpedia’s RDF description of Birmingham (UK) follow a “birthplace” predicate link to a description of the comedian Tony Hancock who was born in the city, from there to BBC broadcasts in which Hancock “starredIn” Traverse RDF links rather than HTML links E.g. Disco hyperdata browser Traverse the web and expose pieces of it in a controlled way, “outline mode” to discover and highlight a pattern of interest and then query for other similar patterns in the web of data E.g. Tabulator Results can be analyzed with conventional presentation methods such as faceted browsers, maps, timelines, etc Track provenance of data while merging data about the same thing from different sources E.g Tabulator, Marbles, FOAFNaut and Fenfire

31. Linked Data Browsers – Marbles Figure 3. The Marbles Linked Data browser displaying data about Tim Berners-Lee. The colored dots indicate the data sources from which data was merged. http://www5.wiwiss.fu-berlin.de/marbles?uri=http%3A%2F%2Fwww.w3.org%2FPeople%2FBerners- Lee%2Fcard%23ihttp://www5.wiwiss.fu-berlin.de/marbles?uri=http%3A%2F%2Fwww.w3.org%2FPeople%2FBerners- Lee%2Fcard%23i (couldn’t reproduce the display above)

32. Linked Data Search Engines and Indexes Browsers navigate Linked Data, Search Engines are where the navigation process begins Provide query capabilities over aggregated data Two categories: 1.Human-oriented search engines 2.Application-oriented indexes 1. Human Oriented Search Engines Falcons and SWSE (Semantic Web Search Engine) Keyboard based search similar interaction as Google and Yahoo Search box into which keywords are entered, returns a list of results Provide a detailed interface to the user that exploits the underlying structure of the data rather than simply providing a link from the search result to the source documents User selects from the results list alongside structured data from the linked entities

33. Human Oriented Search Engine http://ws.nju.edu.cn/falcons/objectsearch/ Object Search for “tim berners lee”

34. Ontology search for “blood”

35. Human Oriented Search Engine SWSE: http://swse.deri.org/

36. Linked Data Search Engines and Indexes cont. 2. Application-oriented Indexes Swoogle, Sindice, Watson (DARQ – federated query) APIs through which Linked Data applications can discover RDF documents on the Web that reference URIs or keywords Applications can query these indexes and receive pointers to relevant documents that can be processed by the application itself Sindice is oriented towards providing access to RDF documents containing instance data Swoogle and Watson are oriented towards finding ontologies that provide coverage of certain concepts relevant to the query

37. Swoogle: http://swoogle.umbc.edu/

38. Search term “blood” finds Ontologies

39. Sindice: http://sindice.org

40. Sindice: search term “blood”, finds Documents

41. Linked Data Domain Specific Applications Domain-specific Applications Revyu, DBPedia Mobile, Talis Aspire, BBC Programmes and Music, and DERI Pipes Domain specific functionality by providing “mashup” of data from various Linked Data sources Revyu Generic review and rating site Consumes Linked Data from the web about films that are reviewed on Revyu to enhance the experience of site users Pulls matching information from other sources such as DBPedia and shows it in the human-oriented pages of the site while maintaining the references to the URIS that may be used by Linked-Data aware applications Information such as director’s name, film posters, books and publications, and external data sets to enhance user profiles with FOAF data

42. Linked Data Domain Specific Applications – cont. DBPedia Mobile Location-aware browser for i-Phpone or other mobile device Oriented towards tourist exploring a city Based on GPS location provides location-centric mashup of nearby locations from DBPedia, associated reviews from Revyu, and photos from Flickr photo- sharing API

43. Linked Data Domain Specific Applications – DBpedia Mobile Figure 5. DBpedia Mobile displaying information about Berlin

44. Linked Data Domain Specific Applications Talis Aspire Resource List Management for university lecturers and students User creates lists through web interface, the application produces and stores RDF triples Items in one list are linked to corresponding items on other lists at other institutions building a web of Scholarly data through non-[linked-data] users BBC Programmes and Music Uses DBpedia and MusicBrainz to connect content about BBC radio and television topics to augment the content with additional data DERI Pipes Similar to Yahoo Pipes Provides mashup platform enabling data sources to be plugged together to form new feeds of data May contain identifier consolidation, schema mapping, RDFS or OWL reasoning and data transformations expressed using SPARQL CONSTRUCT operations or XSLT templates

45. Linked Data Domain Specific Applications – DERI pipes a workbench to produce an output stream of data from 3 sources – http://pipes.deri.org Figure 6. DERI pipes workflow integrating data about Tim Berners-Lee from three data sources.

46. 6. Related Developments (in Research and Practice) Microformats Aim is to extend HTML pages to include structured data Defines a set of formats embedded into HTML via class attributes 2 major differences in the Microformats vs Linked Data RDF serialization 1.Linked Data not limited in the vocabularies that can be used – Microformats are restricted to a small set of vocabularies closely managed by a specific community 2.Data items in HTML via Microformats do not have their own identifiers which prevents assertions across documents and web sites whereas Linked Data URIs are global identifiers that can be used to represent relationships

47. 6. Related Developments (in Research and Practice) Web APIs Amazon, eBay, Yahoo! And Google 1,309 Web APIs and 3,966 mashups based on these APIs APIs are accessed using a wide range of mechanisms and retrieved data is represented using various content formats Difference is that Linked Data is committed to a small set of standard technologies: URIs as identifiers, HTTP as access mechanisms, and RDF as content format Single set of technologies results in ability to use generic data browsers and search engines Most Web APIs don’t assign unique identifiers to data items therefore can’t be linked across different data sources And Mashups are implemented against a set of fixed data sources By contrast, Linked Data can take advantage of connecting different data silos into a single global information space

48. 6. Related Developments (in Research and Practice) Dataspaces System architecture around which research on reference reconciliation, schema matching and mapping, data lineage, data quality and information extraction are unified Offer ‘best-effort’ answers Semantic cohesion is increased over time by different parties providing mappings Web of Data seen as a realization of dataspaces concept on a global scale

49. 6. Related Developments (in Research and Practice) Semantic Web Evolution of human-readable documents to contain more machine- readable semantic information seen as the seeds for what is known as Semantic Web A web of data that can be processed directly or indirectly by machines Web of Data is the goal of Semantic Web Linked Data provides the means by which to reach this goal Building the Web of Data with Linked Data as the foundation may facilitate the reality of intelligent agents and other promises of a more sophisticated Semantic Web vision

50. 7. Research Challenges User Interfaces and Interaction Paradigms Challenge is linking and presenting dynamically linked and potentially unexpected information to a user without unacceptable cognitive overload Example: traditional browsers allow user to move forward and backward in a document-centric resource Similar navigation is expected in a Linked Data browser but instead of within a document, it is moving forward and backwards between entities [of different types], changing the focal point of the application Will need to be able to add/remove resources from view Sindice gives an indication of how such functionality could be delivered, but data sources in the thousands and millions will be a research challenge

51. 7. Research Challenges Application Architectures Semantic Web Client Library and SQUIN have demonstrated that queries can be answered by relying on runtime link traversal Potential problem with scalability of on-the-fly link traversal and federated query Widespread crawling and caching may become the norm Schema Mapping and Data Fusion Retrieving data from distributed sources presents a challenge for how to display the information to the user Most browsers just present the information along side each other Requires mapping of terms from different vocabularies and fusing data about the same entity Data sources can publish correspondences between their local terminology and terminology of related data sources Use W3C recommendations from RDF Schema and OWL termionlogy like owl:equivalentClass, owl:equivalentProperty, rdfs:subClassOf, rdfs:subPropertyOf to publish correspondences

52. 7. Research Challenges Schema Mapping and Data Fusion (cont.) These can be too coarse-grained to perform transformations Need more fine grained schema mappings that support transitive mappings and combining partial mappings Several alignment languages have been presented (Haslhofer 2008 and Euzenat & Scharfee & Zimmerman 2007) as well as rules interchange Format (RIF) Data fusion is the process of integrating multiple data items representing the same real world object Main challenge is resolution of data conflicts where multiple sources provide different values for the same property of an object Distinguishing requirements for data fusion related to Linked Data are the scarceness and uncertainty of quality-related meta-information in order to resolve inconsistencies DERI Pipes and KnoFuss architecture are two protypi8cal systems for fusing Linked Data

53. 7. Research Challenges Link Maintenance Content of Linked Data sources change New entities are added, outdated, changed or removed RDF links updated only sporadically leaving dead links pointing to URIs no longer maintained Can lead to large number of unnecessary HTTP requests by client applications Proposed solutions range form recalculating links at regular intervals (Silk or LinQL) with data sources publishing update feeds or subscriptions, to central registries such as Ping the Semantic Web that keep track of new or changed data items Licensing Applications consuming data need to access specifications of the terms for reusing and republishing the data Creative Commons - a framework for open licensing underpinned by the notion of copyright - others say copyright law doesn’t apply to data Should adopt Open Data Commons Public Domain Dedication and License Further research is needed for interfaces where attribution is required for reuse

54. 7. Research Challenges Trust, Quality and Relevance This is the challenge of ensuring that the data more relevant to the user is identified Content-based, context-based and rating-based techniques that can be used are given in (Bizer & Cyganiak, 2009; Heath, 2008a) PageRank and other algorithms will be important for coarse-grained measures of popularity – as a proxy for relevance or quality – but need to be adapted for the Web of Data Interfaces for how to represent this information is a significant research challenge An “Oh, yeah?” button WIQA and InferenceWeb may be able to contribute to this area by providing explanation about information quality and inference processes used to derive query results

55. 7. Research Challenges Privacy Opportunity to violate privacy when aggregating data from so many distinct sources Likely will require technical and legal means together with higher awareness of users about what to provide in which context Research is being done by Weitzner, 2007 and the TAMI project on information accountability (Weitzner, et al., 2008)

56. Conclusions Linked Data is being adopted by increasing number of data providers Has the potential to enable a revolution in how data is accessed and utilized just as Web brought revolution in how documents were accessed and utilized Developers utilizing Mashups have the challenge of scaling beyond fixed data sources whereas Linked Data is on top of unbounded data sources via standardized access mechanisms

57. Denise’s take: There is potential to build data sources alongside the terminologies also published in RDF to leverage encoded knowledge represented by the ontologies used to annotate caBIG data The technology is still very techie oriented, not accessible to the average Joe Whereas most people can learn HTML or wiki markup pretty easily to publish data and attach a file Have to learn RDF, OWL various different vocabularies for expressing predicates Learn to use the link generation tools, set them up Set up a server to dereference URIs …. Jim will tell us more!

58. EXTRA Slides

59. Uses HTTP redirect status code: “303 See Other” Solution proposed by W3C's Technical Architecture Group in its httpRange-14 resolution [httpRange] httpRange-14 resolutionhttpRange Indicates that the URI is not a web document Directs you to a document about the thing you asked about Avoids ambiguity about the real-world object and the resource that represents it http://www.example.com/id/exampleinc Example Inc., the company http://www.example.com/id/bob Bob, the person http://www.example.com/id/alice Alice, the person Source: http://www.w3.org/TR/cooluris/ [303 URIs for Linked Data]

60. [Relationship between a resource identifier and its documents]

61. [303 URIs for Linked Data] Source: http://www.w3.org/TR/cooluris/

62. [Hash URIs for Linked Data] Hash URI Used for non-document resources Clients strip off the fragment represented at the hash before retrieving the URI The part after the hash cannot be retrieved directly and therefore is not necessarily a document http://www.example.com/about#exampleinc Example Inc., the company http://www.example.com/about#bob Bob, the person http://www.example.com/about#alice Alice, the person Source: http://www.w3.org/TR/cooluris/

63. [Hash URIs for Linked Data] Source: http://www.w3.org/TR/cooluris/

64. Linked Data Applications – Search Engines and Indexes cont. The document web and the data web from one connected navigable space A User may perform an HTML document based query, follow a link into the Web of Data, then follow another link into a different HTML document The search engines might be expected to perform more sophisticated queries but so far that has not proven to be the case [2009] with the exception of Tabulator’s style of query-by-example and faceted browsing SWSE provides access to the underlying data store via SPARQL, but this is suitable for application developers with knowledge of the language rather than a user asking specific questions through human interface