In this position paper, we rst discuss how modern search engines, such as Google, make use of Linked Data spread in Web pages for displaying Rich Snippets. We present an example of the technology and we analyze its current uptake. We then sketch some ideas on how Rich Snippets could be extended in the future, in particular for multimedia documents. We outline bottlenecks in the current Internet architecture that require xing in order to enable our vision to work at Web scale.
? While the author is a liated with Google, this paper is in no way to be seen as an
o cial product strategy, nor as a future roadmap. Opinions are solely the respective
author's.
the presentation layer and that aims to make sense of the data in such a way
that it establishes interoperability between di erent applications [
In this paper, we rst describe the Google Rich Snippet technology as a way of adding more semantics to Web applications, the variety of formats supported, and some rough analysis of its usage (section 2). We then present some ideas of future extension of this technology, in particular for multimedia content (section 3). Finally, we take the initial steps towards a complete overhaul of parts of the current Internet architecture by introducing the thought-experiment of Triple-centric Networking, inspired by Jacobson's Content-Centric Networking (section 4). 2
What are the costs and bene ts of publishing data and semantic markup on the Web4? One needs to carefully examine these issues from di erent perspectives: consumer and publisher. We are starting to see a number of domains, such as the Public Sector Information/eGov area, Life Sciences, eCommerce, and others to not only be aware of the bene ts of publishing Linked Data, but start to exploit it on a large scale. In this section, we present XHTML code samples with embedded RDFa mark-up and we show how Google uses this information in its Rich Snippets technology. 2.1
Google introduced Rich Snippets on May 12, 2009 as a means of displaying
structured data in search result pages with the objective of highlighting the
searchedfor properties to the user in a visually outstanding way [
Multiple semantic mark-up formats are supported by Google (RDFa,
Microformats, Microdata), the search company taking a very practicable approach: A
lot of previous work on structured data has focused on debates around encoding.
Even within Google, we have advocates for microformat encoding, advocates for
various RDF encodings, and advocates for our own encodings. But after working
on this Rich Snippets project for a while, we realized that structured data on
the web can and should accommodate multiple encodings: we hope to emphasize
this by accepting both microformat encoding and RDFa encoding. Each encoding
4 http://lab.linkeddata.deri.ie/2010/star-scheme-by-example/
5 http://microformats.org/wiki/Main_Page#Specifications
6 http://www.google.com/support/webmasters/bin/request.py?contact_type=rich_snippets_
-feedback
has its pluses and minuses, and the debate is a ne intellectual exercise, but
it detracts from the real issues [
The Rich Snippets technology support various vocabularies. For example, details about an o ering such as the ticket booking of an upcoming event can be marked up in the body of a Web page in order to help understanding the location, schedule, price or reviews of the event. The code example in Figure 1 illustrates the mark-up for an event at a certain business location.
The example begins with a namespace declaration using xmlns. In the rst line, typeof="v:Event" indicates that the marked-up content describes an Event. The dimensions that composed the event (description, type, starting time) are described with properties. The property name is pre xed with v: (<span property="v:description">). Google does not display information that isn't visible to the user, with a few exceptions. Geo information (latitude and longitude of the location) can be included in the HTML markup. Typically, this information is not visible on a Web page about an event, but providing it can help ensure that the location is accurately mapped.
Figure 2 shows how a search result using this semantic markup is then displayed on Google search result page. 2.3
Goel et al. have compiled some statistics with regards to semantic mark-up on
the Web in June 2010 [
Further analysis of the dataset crawled has shown a number of pitfalls: incorrect labeling (e.g. marking up the date of an event as part of the event description), or incorrect inclusion of unrelated words in the structured mark-up (e.g. marking up \written by John Doe" rather than just \John Doe" as value of the property v:reviewer). Furthermore, they observe a general confusion with what parts of a document should be marked up at all. Although some web pages include RDFa event markup, none of them are used by the Rich Snippet technology as of today. 7 http://googlecustomsearch.blogspot.com/2009/05/enabling-rich-snippets-in-custom-search.
html 8 http://developer.yahoo.com/search/boss/boss_guide/ <div xmlns:v="http://rdf.data-vocabulary.org/#" typeof="v:Event"> <a href="http://www.example.com/events/poisel_offenback.hmtl" rel="v:url" property="v:summary">Philipp Poisel in Offenbach</a> <span property="v:description">See Philipp Poisel in Offenbach</span> When: <span property="v:startDate" content="2011-01-16T19:00-01:00">
Jan 16, 7:00PM</span> <span property="v:endDate" content="2011-01-16T21:00-01:00"> 9:00PM</span> Where: <span rel="v:location"> <span typeof="v:Organization"> <span property="v:name">Capitol</span>, <span rel="v:address"> <span typeof="v:Address"> <span property="v:street-address">Kaiserstra e 106</span>, <span property="v:locality">Offenbach am Main</span>, </span> </span> <span rel="v:geo"> <span typeof="v:Geo"> <span property="v:latitude" content="50.10945"></span> <span property="v:longitude" content="8.76579" ></span> </span> </span> </span> </span>
Category: <span property="v:eventType">Concert</span>
</div>
Improving how search results are displayed in snippets preview involve a huge
market. Tickets aggregator Web sites such as Giga-Music.de place so-called
afliate links to the nal ticket vendors. These sites, very often, make their living
from accumulating useful metadata around a certain event, and then, provide
only links to vendors in order to nally get paid, e.g. on a pay-per-click model.
In [
Search engines serve mainly as entry points to the Web. Let us imagine a user wants to see a concert of the German singer and songwriter Philipp Poisel. A straightforward query would be \philipp poisel konzerte9". At the time of writing, this search results in: the concerts section of the artist's o cial Website as the rst result10, a couple of fan club sites11, some concert review sites12, and some ticket aggregator sites13.
In the following, we purely focus on how Linked Data could enrich the search experience, for example, for a Web search for Philipp Poisel concerts. We anticipate a huge uptake of semantic mark-up by Web site operators over the coming 9 \konzerte" is the German word for \concerts". 10 http://www.philipp-poisel.de/termine/ 11 E.g. http://www.philipppoiselfanclub.de/?tag=konzert 12 E.g. http://www.ciao.de/Philipp_Poisel_Konzert_Tickets__8025011 13 E.g. http://www.giga-music.de/philipp-poisel-konzerte-live-tour/ months. Especially, new business-related vocabularies such as the Tickets Ontology[5], are expected to see broader and broader usage and implementation. In the following, we assume that ticket vendors had implemented the Tickets Ontology on their Web pages. Therefore, we consider a couple of pages with the following triples (for the sake of clarity we omit the necessary pre xes and simplify the xsd:dateTime format): foo:ticket a tio:TicketPlaceholder ; rdfs:label "Tickets for Philipp Poisel"@en ; tio:accessTo <http://data.events.example.org/123> . foo:ExampleTicketVendor gr:offers foo:offer . foo:offer a gr:Offering ; gr:name "Tickets for Philipp Poisel in Bremen"@en ; gr:description "Philipp Poisel in Bremen"@en ; gr:includes foo:ticket ; gr:hasBusinessFunction gr:Sell ; gr:hasPriceSpecification [a gr:UnitPriceSpecification ; gr:hasCurrency "EUR"@en ; gr:hasCurrencyValue "24.10"^^xsd:float ; gr:validThrough "2010-11-11T23:59"^^xsd:dateTime].
Each ticket instance has access to a tio:Event, which subclasses a lode:Event[8], an event:Event, and a dul:Event, and where the particular concert event dates are described. We assume that the Web site also implements v:Event. This would allow for comparative Rich Snippets. The obvious next step would be to bring the social experience into play, granted the user has given access to her social graph . This would mean to carry part of the \Facebook experience" right into the search experience. Prior research by Troncy et al. has shown that users generally start an event search on a general search engine, and that the decision criteria whether or not to attend an event is often dependent on whom of the user's friends plan to attend [10, 9].
An entirely di erent class of extended Rich Snippets could be based on multimedia semantics in order to provide richer video search results. We believe that there is high potential for semantically annotated multimedia content to improve content search. In Figure 4, we show a mock-up of a person highlighted, which could be based on media fragment URIs. Such media fragment URI could look like: The components of this URI are rst a temporal dimension (t=428,434), which selects seconds 428 to 434 of the whole video, and then a spatial dimension (xywh=150,60,50,70 and xywh=240,50,50,70), which creates two bounding boxes at the x, y parameters with a width w and a height h.
Fig. 4. Sketch of an extended Rich Snippet featuring semantically highlighted video preview (still frame or moving images), actor information, and in-video links.
The vision of extended Rich Snippet outlined above features information from more than just one data source which is di erent from today's Rich Snippets where the content is exclusively determined by the information in one particular Web page. It is obvious that in order to combine information coming from various data sources, an information sharing mechanism must be established. In the following, we sketch a thought-experiment derived from Content-centric Networking introduced by Jacobson [7]. In Content-centric Networking, there are two notions of packages involved: Interest and Data packages. Interests get broadcast by consumers, and as soon as a node can satisfy an interest, it responds with the data. Otherwise, it rebroadcasts the interest. The main advantage over common host-based networking is that data packages are not only exclusively thought for the initially interested node, but can be shared between nodes with common interests. A certain piece of information satis es an interest if the content name in the interest package is a pre x of the content name in the data packet. Applied to RDF triples, this could mean that the content name would correspond to the subject. Figure 5 is adapted from Figure 2 in [7] and illustrates how the interest and data packages could look like if we applied the principle of Content-centric Networking to Triple-centric Networking.
Experiments by Jacobson et al. have shown that Content-centric Networking is useful when many parties are interested in the same content. This is de nitely the case if we think Web scale triple propagation for popular Web pages. One of the potential bottlenecks of rolling out multi data sources Rich Snippets could thus be avoided by moving triple propagation to this new Triple-centric Networking layer. We are at the early stages of this thought-experiment and have not carried out any experimentation to justify our assumption. However, we believe that there is a potential improvement to the host-to-host networking infrastructure that dominates today's Internet tra c. 5
We have shown why and how Rich Snippets formats are currently supported by Google and we have provided code samples in RDFa. We have then brie y discussed the potential business impacts of Rich Snippets to Web site operators. It has become visible that Rich Snippets are a very sensible element in the Linked Data value chain due to the high visibility and the con rmed change of user click-through behavior. We have outlined potential extensions to Rich Snippets, driven by a concrete use case of an event-based Web search. We have interlinked the social graph of a user with common event-related data in the Linked Data cloud. In our mock-up, we neglected the business impact of Rich Snippets entirely. However, it is obvious that for the online ticket search example, the decision what ticket vendor to include, and what vendor to exclude from the vendors shown in the Rich Snippets is a crucial one.
The suggested addition of a Linked Data layer [
The research leading to this paper was partially supported by the project AAL2009-2-049 \Adaptable Ambient Living Assistant" (ALIAS) co-funded by the European Commission and the French Research Agency (ANR) in the Ambient Assisted Living (AAL) programme, and by the projects FP7-216444 \Peer-topeer Tagged Media" (Petamedia), FP7-248296 \I-SEARCH" and FP7-256975 \LOD Around-The-Clock" (LATC) Support Action. 5. M. Hepp. Tickets Ontology v1.0, November 17, 2010. http://purl.org/tio/ns. 6. I. Hickson. HTML Microdata. W3C Working Draft, October 19, 2010. http: //www.w3.org/TR/microdata/. 7. Jacobson, D. Smetters, J. Thornton, M. Plass, N. Briggs, and R. Braynard. Networking Named Content. In 5th ACM International Conference on Emerging Networking Experiments and Technologies (CoNEXT'09), Rome, Italy, 2009. 8. R. Shaw, R. Troncy, and L. Hardman. LODE: Linking Open Descriptions Of Events. In 4th Asian Semantic Web Conference (ASWC'09), pages 153{167, Shanghai, China, 2009. 9. R. Troncy, A. Fialho, L. Hardman, and C. Saatho . Experiencing Events through User-Generated Media. In 1st International Workshop on Consuming Linked Data (COLD'10), Shanghai, China, 2010. 10. R. Troncy, B. Malocha, and A. Fialho. Linking Events with Media. In 6th International Conference on Semantic Systems (I-SEMANTICS'10), Graz, Austria, 2010.