=Paper=
{{Paper
|id=None
|storemode=property
|title=Schema.org for the Semantic Web with MaDaME
|pdfUrl=https://ceur-ws.org/Vol-1026/paper3.pdf
|volume=Vol-1026
|dblpUrl=https://dblp.org/rec/conf/i-semantics/VeresE13
}}
==Schema.org for the Semantic Web with MaDaME==
https://ceur-ws.org/Vol-1026/paper3.pdf
Schema.org for the Semantic Web with MaDaME
Csaba Veres, Eivind Elseth
Department of Information Science and Media Studies,
Postbox 7802, University of Bergen, 5020 Bergen, Norway
csaba.veres@infomedia.uib.no, eivind.elseth@student.uib.no
Abstract. Schema.org is a high profile initiative to introduce structured markup
into web sites. However, the markup is designed for use cases relevant to search
engines, which limits their general usefulness. MaDaME is a tool to help web
developers to annotate their web pages with schema.org annotations, but in addi-
tion automatically injects semantic metadata from SUMO and WordNet. It is un-
like previous tools in that it assumes no knowledge of the metadata standards.
Instead, users provide disambiguated natural language terms, and the tool auto-
matically picks the most appropriate metadata terms from the different vocabu-
laries.
Keywords: schema.org, wordnet, semantic web, markup, search
1 Introduction
Schema.org was launched on June 2, 2011, under the auspices of a powerful consorti-
um consisting of Google, Bing, and Yahoo! (they were subsequently joined by Yan-
dex). They established the http://schema.org web site whose main purpose is to docu-
ment an extensive type schema which is meant to be used by web masters to add struc-
tured metadata to their content. In a sense schema.org provides extended semantics for
rich snippets1 with the motivation that markup can be used to display more infor-
mation about web sites on the search page which may result in more clicks, and per-
haps higher rankings in the long run.
The schema was designed specifically for the use cases developed by the search
engines, and both the semantics and preferred syntax reflect that choice. In terms of
semantics, the schema has some non-traditional concepts to fulfill its role. For exam-
ple there is a general class of Product but no general class for Artifact. There are also
odd property ascriptions from the taxonomy structure, so, for example, Beach has
openingHours and faxNumber. These oddities exist because, we are told, they reflect
what people are predominantly looking for when they perform a search. In terms of
syntax, there is a very strong message that developers should use the relatively new
Microdata format, designed specifically for the schema, rather than the vastly more
popular RDFa web standard [1]. The choice is dictated by simplicity, because Micro-
data has just those elements required for the schema. But this choice is unfortunate
1
http://goo.gl/RAJy8
11
Schema.org for the Semantic Web with MaDaME
because it makes metadata from schema.org incompatible with many other sources of
metadata like Facebook’s OGP.2 [2] lists five key reasons why RDFa Lite 1.1 should
be the preferred syntax over Microdata. RDFa is feature equivalent to Microdata, and
it is supported by all major search crawlers including Facebook, while Microdata is
not. For the purposes of expressing schema.org, RDFa is no more complex than Mi-
crodata. But most importantly from the perspective of general semantic markup, RDFa
is designed to naturally mix vocabularies while Microdata makes it much more diffi-
cult to do so. Thus if annotating web pages with multiple vocabularies is the desired
goal, then RDFa Lite 1.1 is the best choice.
MaDaME (Meta Data Made Easy) is a markup tool developed for two specific
purposes. First, it must to help web developers who were not familiar with the sche-
ma.org to mark up their web sites as easily as possible. This is important because the
idiosyncratic nature of the schema can make concepts hard to navigate. It is especially
important if a web developer wants to mark up a site for which there is no existing
type in schema.org. For example a web master might be designing a web site about
caves for tourists to visit, but schema.org does not have a type for cave. We wanted to
help developers find the best markup in these cases, without requiring them to study
the schema itself. The second important motivator was to make the markup episode as
fruitful as possible, since it is not easy to motivate people to provide structured data
about their web site. This means the markup should be useful in as many use cases as
possible. We achieve this by producing RDFa markup and mixing different vocabular-
ies to describe the same object. While there are existing efforts to provide tool support
for schema.org markup, including a tool from Google, 3 all of them require some
knowledge of the schema, and none of them provide rich markup for a more general
semantic web.
2 MaDaME
MaDaME has at its core a mapping file between WordNet word senses and sche-
ma.org types. WordNet can for our purposes be regarded as a comprehensive electron-
ic dictionary which defines word senses through numerous relationships to other
words [3]. Web developers simply look up the word which expresses the content of
their site, and they are given the best matching schema.org markup. Obviously not all
words will have direct mappings to schema.org, so we also have an algorithm to infer
the best match for those.
To import a page into the web app the user will write the URL of the web site he
wants to mark up into the URL input field. The page will then be loaded into the web
app after some preprocessing. The preprocessing consists of commenting out scripts
and iframes which might not run correctly. The user then selects words, phrases, or
images to tag by highlighting them on the page. When a word item has been highlight-
ed, its possible senses in WordNet are retrieved. The user picks one of these senses by
clicking on it. In fig. 1 we can see the word ridge highlighted, and the corresponding
disambiguation options. The sense the user picks is sent back to the server for map-
2
http://ogp.me
3
http://goo.gl/7DGr5D
12
Schema.org for the Semantic Web with MaDaME
ping to schema.org, as well as a selection of other ontologies. So far we have only
implemented SUMO [4] and WordNet itself. The Schema.org mappings can be further
refined by filling out the properties defined by the schema, using a popup form.
When the users finish marking up the document they are given a link to a newly
created webpage containing their original page plus the meta data they have created. In
most cases where the web site is simple HTML there will be no need to manually
modify any code. From here they can save the document and upload it to their own
server.
Fig. 1. A screenshot of MaDaME with options for ridge on the left of the screen
All of the markup is in the RDFa Lite 1.1 syntax, which is the current W3C rec-
ommendation, 4 and has the necessary features to handle multiple namespaces and
multiple types elegantly.
The algorithm for finding markup for the selected senses is in two stages. The first
stage is to build an extended tree of WordNet senses. This is done by using a perl li-
brary (the WordNet::QueryData library from CPAN) which is capable of querying the
WordNet database. We have written a script that when given a WordNet sense will
find the hypernyms of the sense (more general senses), and all the hyponyms (more
specific) of these ancestor nodes. We call this the mapping tree, which intuitively
contains all the words in the semantic neighbourhood of the original word.
In the second stage we find mappings for the user selected synsets. If a direct map-
ping to schema.org exists then this is simply added to the markup. For novel words we
4
http://www.w3.org/TR/rdfa-lite/
13
Schema.org for the Semantic Web with MaDaME
use mappings for the closest available related sense from the mapping tree which does
have a direct mapping. We tried several versions of the mapping algorithm, and the
most successful one turned out to be a simple depth-first traversal of the mapping tree
until a sense is found with a direct mapping to the schema. For a simple example, con-
sider the concept ridge which is not represented in schema.org. The correct sense of
wn:ridge has the hypernym wn:geological_formation, which has a direct mapping to
schema:Landform. Therefore ridge is marked up as schema:Landform. SUMO has
direct mappings for a very large number of WordNet senses and ridge has a corre-
sponding mapping in SUMO as sumo:UplandArea, so the concept ridge would acquire
mappings schema:Landform as well as sumo:UplandArea. More generally, any vo-
cabulary that is mapped to WordNet could be used to provide metadata. In future re-
leases we plan to provide facilities for advanced users to incorporate their own map-
ping files to an ontology of their choice.5
3 Results
We performed an automatic evaluation of 4350 random nouns in WordNet to see how
they mapped to schema types, by measuring the average depth of the mapped type in
the schema.org taxonomy. The result was a somewhat disappointing 0.689, which
means that most words were mapped to schema:Thing or one of its immediate special-
isations.
To test how this compares to real world usage we sampled a set of five web sites
that had used schema.org markup. We ended up with a restaurant review from the
Telegraph, a tour operators customer feedback page, a tourist agency home page, the
home page of a marketing company and a movie review sites review of a film. When
we manually added markup by selecting key words in the text we achieved 100%
agreement. While this is clearly a small study, it does suggest that the schema.org
markup we will see “in the wild” will represent concepts from the top nodes of the
type hierarchy. The relatively shallow mappings may be a reflection of the schema
itself, rather than a criticism of our mapping algorithm.
4 Related Work
There are existing approaches for annotating web pages with semantic markup, espe-
cially schema.org. These can broadly be categorised as manual or automatic annota-
tion tools.
The schema.rdfs.org web site links to a number of publishing tools 6. The two ma-
jor form-based tools, Schema Creator and Microdata Generator, both provide a forms
based interface for entering detailed properties, not unlike the MaDaME interface.
However in these tools the web author must find the appropriate schema types by
5
The tool can be tried at http://csaba.dyndns.ws:3000.
6
http://schema.rdfs.org/tools.html
14
Schema.org for the Semantic Web with MaDaME
browsing a sub set of the most common types that are presented in these tools. They
both differ from our approach because they expect the author to make decisions about
which schema types to use. Similarly, major content management platforms like Dru-
pal, Joomla!, WordPress and Virtuoso provide mechanisms for adding schema.org
types to their content.
Amongst automatic annotation tools, [5] presents a tool that can add schema.org
types automatically, but only to web pages about patents. Their approach uses underly-
ing domain knowledge to extract key terms and a patent knowledge base to generate
structured microdata markup for web pages. It remains to be seen if this approach
could scale to web sites in general.
5 Conclusion
Schema.org is a promising initiative from the search engines in that it exposes struc-
tured metadata to a vast new audience of web developers. However, this requires some
learning of the syntax and vocabulary of the particular markup, which could limit the
breadth of metadata that will appear from web developers. MaDaME is a tool that
helps web masters use the schema because it removes the requirement to learn a new
vocabulary and syntax, while providing the necessary markup. The markup can be
extended to other proprietary standards like Facebook’s OGP, so web sites could be
annotated with both standards at no extra effort. But we see MaDaME’s most im-
portant contribution as one to the semantic web effort because it piggybacks on the
major search engine backed initiative, to include markup from popular ontologies that
can be used for diverse semantic applications.
References
1. P. Mika and T. Potter, “Metadata Statistics for a Large Web Corpus,” WWW2012 Work-
shop on Linked Data on the Web (LDOW '12), Lyon, France, 16-Apr-2012. Online Availa-
ble: http://ceur-ws.org/Vol-937/ldow2012-inv-paper-1.pdf [Accessed: 11-Jul-2012].
2. M. Sporny, “Mythical Differences: RDFa Lite vs. Microdata | The Beautiful, Tormented
Machine,” manu.sporny.org. Online Available: http://manu.sporny.org/2012/mythical-
differences. [Accessed: 17-Jul-2013].
3. G. A. Miller, “WordNet: a lexical database for English,” Communications of the ACM, vol.
38, no. 11, Nov. 1995, pp. 39–41
4. ௗI. Niles and A. Pease, “Towards a Standard Upper Ontology,” Proceedings of the Interna-
tional Conference on Formal Ontology in Information Systems (FOIS '01), ACM, 2001, pp.
2–9.
5. A. Norbaitiah and D. Lukose, “Enriching Webpages with Semantic Information,” Proc.
Int’l Conf. on Dublin Core and Metadata Applications 2012, Sep. 2012, pp. 1–11
15