Linking XBRL Financial Data: Linking Enterprise Data

Linking XBRL Financial Data

Roberto García and Rosa Gil

Abstract One of the main ways of populating the Web of Data is by translating

existing data sources. One interesting candidate for this approach is data based

on the eXtensible Business Reporting Language (XBRL), a standard for business

and ﬁnancial reporting. Many institutions are making available or requiring data in

this format, e.g. the U.S. Securities and Exchange Commission (SEC) through the

EDGAR program. However, XBRL data is loosely interconnected and it is difﬁcult

to mix and query it. Our contribution is a translation from XBRL ﬁlings to Semantic

Web technologies, which we have applied to more than 1000 ﬁlings obtaining more

than 2 million triples. The resulting semantic data is easier to integrate and cross

query. Moreover, it can be interconnected with the rest of the Web of Data in order

to extract its full potential.

1 Introduction

The main way to populate the Web of Data is by translating existing data sources.

The motivation to do so is that usually this data is not offering its full potential

because it is isolated, i.e. not connected to other external pieces of data that enrich

them. It might even be the case that the data is loosely interconnected internally.

Most of the time this is due to the fact that the technological solutions used to publish

that data do not make it easy to interconnect it internally and to other external data

sources.

Business reporting is a domain where the need for a common data format for re-

ports has already been identiﬁed. XBRL (eXtensible Business Reporting Language)

Roberto García

Universitat de Lleida, Jaume II, 69. 25001 Lleida, Spain

Rosa Gil

Universitat de Lleida, Jaume II, 69. 25001 Lleida, Spain

103

104

Roberto García and Rosa Gil

is an XML language intended for modeling, exchanging and automatically process-

ing business and ﬁnancial information. XBRL is being deployed in many different

scenarios, especially thanks to the support of some regulators and government agen-

cies. For instance, there is the EDGAR program promoted by the U.S. Securities and

Ex-change Commission (SEC). It performs automated collection, validation, index-

ing, acceptance and forwarding of submissions by companies and others who are

required by law to ﬁle forms with the SEC.

It has evolved from a voluntary program and now there is a mandate for a three

years phase-in schedule starting 2009 with companies with public ﬂoat over $5

billion (approximately 500 companies) and ending 2011 with all companies ﬁling

to the SEC doing so using XBRL. Moreover, the Government Information Trans-

parency Act will require federal agencies to collect their data in a uniform, search-

able format using XBRL thereby simplifying mandatory ﬁnancial reporting for com-

panies that receive federal funds.

However, we have observed limited support for cross analysis of ﬁnancial in-

formation in XBRL tools and applications, as it is detailed in Section 1.2. This is

not just among data based on different accounting principles, which are represented

in XBRL using taxonomies. It even happens when comparing ﬁlings for different

companies based on the same taxonomies or ﬁlings for the same company based on

different versions of the taxonomies.

We argue that this limitation is inherited from the technologies underlying

XBRL, especially XML. XML takes a document oriented approach, where each

document presents a tree structure. This makes it difﬁcult for XML-based tools to

provide functionalities that blur this separation into documents and that overcome

the limitations of a tree structure when mashing-up data from different sources.

Moreover, XBRL does not provide formal semantics that might help to integrate

different taxonomies using logic reasoners.

In any case, the integration of data contained in XBRL into comparable informa-

tion is a strong requirement for the analysis of business and ﬁnancial information at

the global level. This might increase the efﬁciency and effectiveness of the decision

making processes relying on this kind of information. For instance, bankruptcy pre-

diction and other tasks related to the assessment of the solvency of a ﬁrm, a business

sector or set of interrelated companies.

Many have already pointed to this issue and propose Semantic Web technologies

as a natural choice for data integration [1] and, in this concrete case, for XBRL data

integration, cf. the related work in Section 1.2 or the W3C Workshop on Improving

Access to Financial Data on the Web 1 . However, this is not enough, the Semantic

Web provides the technologies for data integration but some principles are required

that facilitate Web-wide deployment of highly interlinked XBRL data. Linked Data

[2] provides these principles to publish data in the World Wide Web in a way that

helps making it easily discoverable through the links that connect it to other pieces

of data.

1 Program of the W3C Workshop on Improving Access to Financial Data on the Web,

http://www.w3.org/2009/03/xbrl/program.html

Linking XBRL Financial Data

105

Despite these beneﬁts, currently, ﬁnancial and business data is being produced

using XBRL and it seems that more and more XBRL data is going to be available in

the future. XBRL is been promoted by regulators and government agencies like the

SEC, as it has been shown before, but also other bodies like the European Union or

the Spanish securities commission [3].

Consequently, our opinion is that the best short term approach in order to get

ﬁnancial and business data to the Semantic Web is not to propose and alternative

language based on Semantic Web technologies, but to apply methods to map exist-

ing XBRL to semantic metadata. This also seems the best option in the short and

midterm to populate the Web of Data with business information.

The rest of this chapter is organised as follows. The next subsections introduce

the structure of XBRL, then the related work is presented followed by the descrip-

tion of the our contribution in Section 2. It is based on the XML Semantics Reuse

Methodology. The ﬁrst step is to map the XML Schemas that structure XML data

to OWL ontologies using the XSD2OWL mapping. Then, the second step is to map

XML data to RDF using the XML2RDF mapping.

Once our approach has been presented, the results of the previous mappings are

shown in Section 3. They are a set of OWL ontologies for the main XBRL tax-

onomies used in the EDGAR program. Based on these ontologies, it has been pos-

sible to map all the EDGAR instance documents from XML based on these tax-

onomies to RDF based on the resulting ontologies.

From these ontologies and semantic data, it has been possible to establish some

mechanisms, facilitated by Semantic Web technologies, that enrich the dataset with

additional links. First, some links to external datasets of the Web of Linked Data.

Second, internal links that integrate the different ﬁlings by aligning the ontologies

they use.

We are currently starting to evaluate this semantic dataset, as it is detailed in

Section 4. It is compared to similar undergoing initiatives and it has been made

publicly available for querying and browsing through a Web user interface. Finally,

in Section 5, the conclusions and future work are presented. The main conclusion

is that though RDF data through semantic queries and integration primitives offers

a new range of possibilities; it already lacks enough expressive power to substitute

XBRL, as it is explained in the conclusion.

We think that the best approach, for the moment, is to combine both approaches

and transform XBRL data to semantic form in order to facilitate cross-querying and

semantic integration, while keeping the original data in order to beneﬁt from speciﬁc

XBRL services. Consequently, we concentrate now our future work in completing

the mapping from XBRL to Semantic Web, to provide integration facilities at the

taxonomy level and to enrich the links of the resulting semantic dataset to other ones

in the Web of Linked Data.

106

Roberto García and Rosa Gil

1.1 XBRL

XBRL is based on two kinds of documents, instance documents and taxonomies.

In-stance documents report business facts and point to a set of taxonomies, which

deﬁne the meaning of these facts, e.g. under what accounting principles they hold,

what other facts they related to or what kind of things do they refer to.

1.1.1 Instances

More concretely, a XBRL instance document contains business Facts. An example

of a Fact could be “sales in the last quarter”. If the Fact is simple valued, like “the

long term debt is 350,000” whose value is just a number, it is called Item. If the

Fact has a more complex value, like “for the preferred stock , the preferred stock par

value per share is 0 and the preferred stock shares authorized is 2000”, it is called

a Tuple.

Items are represented in XBRL as a single XML element with the value as its

con-tent while Tuples are represented by XML elements containing nested Items or

Tuples, i.e. subelements.

However, facts are not isolated entities and it is not enough to provide their val-

ues, it is also necessary to contextualize them. Consequently, four more entities are

introduced in the XBRL model:

• Context : it deﬁnes the entity (e.g. company or individual) to which the fact ap-

plies, the period of time the fact is relevant and an optional scenario. The period

of time can have zero length for instance and its value is based on ISO 8601 for

date and time values. Scenarios provide further contextual information about

the facts, such as whether the business values reported are actual, projected or

budgeted. Contexts are referenced from Facts using the “contextRef” attribute,

which speciﬁes that the given Fact is valid for the entity, period and scenario

deﬁned in the Context.

• Unit : it deﬁnes a unit of measure, such as “USD” or “shares”. They are refer-

enced from Facts using the “unitRef” attribute, which speciﬁes that the numeric

or fractional value of the Fact is based on that unit of measure. Complex units

can also be deﬁned, like “USD per share”. Currency units are based on ISO

4217.

• Reference : The kinds of facts under consideration are deﬁned by taxonomies,

which specify their meaning in the context of some accounting principles or

purpose, e.g. Facts relevant for banking and savings institutions. These kinds of

facts are then used in instance documents in order to specify actual values for

them. However, they are linked to their deﬁnition in the taxonomies, typically

through schema references, in order to be able to retrieve their meaning.

• Footnote : it contains some additional support content and it is associated to

Fact using XLink 2 .

2

XLink, http://www.w3.org/TR/xlink/

Linking XBRL Financial Data

107

Listing 1 shows part of an instance document from the EDGAR program that

contains a Context element which deﬁnes a company, a time period and the scenario

unaudited. Then, there is a Fact that holds in that context. The Fact references the

Context and the value unit, while their content is the actual numeric value for that

fact.

1.1.2 Taxonomies

The other kind of XBRL document represents taxonomies. A taxonomy deﬁnes a

hierarchy of concepts, basically kinds of Facts, and captures part of their intended

meaning. In XBRL there is a set of base taxonomies that deﬁne the core concepts

and other ones that extend them in order to particularize these concepts for concrete

accounting principles, application domains, etc. Additionally, it is possible to extend

existing taxonomies and accommodate them to particular needs.

Taxonomies are based on XML Schemas, which provide the taxonomy build-

ing primitives and the extension mechanisms. Moreover, there are also “linkbases”,

which allow establishing links beyond the tree structure of a taxonomy by virtue of

their use of XLink:

• Schemas deﬁne concepts that are instantiated as Items or Tuples, depending on

their complexity, in the instance documents. They are based on XML Schema

elements (xsd:element). A concept deﬁnition provides the fact name, whether

it is a tuple or an item and its value data type (such as monetary, numeric,

fractional or textual).

• Linkbases deﬁne links from concepts in a taxonomy to labels, pieces of content

or to other concepts. The XBRL 2.1 speciﬁcation deﬁnes ﬁve different kinds of

linkbases.

– Label Linkbase : set of links that provides human readable strings for con-

cepts, potentially in multiple languages.

– Reference Linkbase : these links associate concepts with citations of some

body of authoritative literature.

– Calculation Linkbase : these are links that associate a set of values of con-

cepts in taxonomies with a mathematical calculation that must be checked

for consistency, for instance that a set of concepts with percentage values

sum up 100%.

– Deﬁnition Linkbase : it provides semantic relations between concepts like

is-a, whole-part, etc.

– Presentation Linkbase : This linkbase associates concepts with other con-

cepts so that the resulting relations can guide the creation of a user interface,

rendering, or visualisation.

Listing 1 Context and facts examples from an EDGAR ﬁling

< c o n t e x t i d = ‘ ‘ From20080301 − T o 2 0 0 8 0 5 3 0 E n t e r p r i s e S o l u t i o n s U n a u d i t e d ” >

108

Roberto García and Rosa Gil

< e n t i t y >

< i d e n t i f i e r schem e = ‘ ‘ h t t p : / / www. s e c . gov / CIK” >

796343

< / i d e n t i f i e r >

< / e n t i t y >

< p e r i o d >

< s t a r t D a t e > 2008 − 03 − 01 < / s t a r t D a t e >

< e n d D a t e > 2008 − 05 − 30 < / e n d D a t e >

< / p e r i o d >

< s c e n a r i o >

< a d b e : U n a u d i t e d / >

< / s c e n a r i o >

< / c o n t e x t >

< a d b e : E n t e r p r i s e S o l u t i o n s R e v e n u e d e c i m a l s = ‘ ‘ − 6”

c o n t e x t R e f = ‘ ‘ From20080301 − T o 2 0 0 8 0 5 3 0 E n t e r p r i s e S o l u t i o n s U n a u d i t e d ”

u n i t R e f = ‘ ‘USD” > 54400000 < / a d b e : E n t e r p r i s e S o l u t i o n s R e v e n u e >

1.2 Related Work

The U.S Securities and Exchange Commission (SEC) offers some online tools that

al-low interacting with the data available in XBRL form. There is a tool called In-

teractive Financial Reports that allows viewing and charting companies ﬁnancial

information. It also provides some functionality that allows comparing different ﬁl-

ings and different companies, though it is hard to use and prone to even the slightest

differences between the compared ﬁling facts, even when there is just a name change

for facts from ﬁlings of the same company.

There is also the Financial Explorer 3 , which presents company ﬁnancial data

through very informative diagrams. In this case, it is just possible to show data from

one company at a time. Finally, there is the Executive Compensation tool, which

al-lows comparing just two facts, Public Market Capitalization and Revenue, across

all ﬁled companies.

Apart from the SEC tools, there are some other XBRL tools, most of them pro-

prietary and with quite high licensing cost. Among them, the Fujitsu XBRL Tools 4

should be highlighted because they are one of the most popular tool sets and it is

available for XBRL Consortium members and academic users. The tools comprise

taxonomy and instance editors, viewers and validators.

The most powerful tool in this set, though still in beta and with many usabil-

ity problems, is the Instance Dashboard. This application can consume multiple

instance documents and, by specifying a base taxonomy, users can perform some

comparison analysis, though limited to facts in the taxonomy that appears in all the

ﬁlings.

3

SECs Financial Explorer, http://209.234.225.154/viewer/home/

4

Fujitsu XBRL Tools, http://www.fujitsu.com/global/services/software/interstage/xbrltools/

Linking XBRL Financial Data

109

As it can be noted from the previous analysis, the main limitation of XBRL tools

is their limited support for cross analysis of ﬁnancial information, not just among

data based on different taxonomies, even when comparing ﬁlings for different com-

panies based on the same taxonomies.

This limitation is inherited from the technologies underlying XBRL, especially

from XML. XML takes a document oriented approach, where each document

presents a tree structure. This makes it difﬁcult for XML-based tools to provide

functionalities that blur this separation into documents and that overcome the limi-

tations of a tree structure when mashing-up data from different sources.

Consequently, Semantic Web tools are being considered by people like Charles

Hoffman, the father of XBRL: “ This ﬁeld (W3C semantic standards) is rich with

possibilities and stands as the next logical step in the natural progression of infor-

mation technology to seek a higher value proposition ” (emphasis added) [4].

This interest is materializing, and the combination of XBRL and the Semantic

Web has been receiving some attention in different blogs 56 , mailing lists and web

groups 7 . However, it is difﬁcult to ﬁnd concrete results that put into practice Seman-

tic Web technologies in the XBRL ﬁeld.

Moreover, most of these results are speciﬁc for some parts of XBRL. For in-

stance, there is an ontology about ﬁnancial information based on XBRL that is spe-

ciﬁc for investment funds [5] and, though it is generated using a generic XBRL

taxonomy to OWL ontology algorithm, there is not and equivalent tool that maps

generic XBRL instance data. There is also another tool that maps quarterly and

semester accounting information submitted to the Spanish securities commission

(CNMV) to Semantic Web technologies [3].

Moreover, both approaches are based on procedural code specially developed in

order to extract speciﬁc patterns from the XBRL data. Consequently, they are difﬁ-

cult to scale to the whole XBRL speciﬁcation and sensible to minimal changes in it.

We propose an approach that, instead of directly processing XBRL data, takes proﬁt

from the fact that it is expressed using XML and speciﬁed using XML Schemas.

OpenLink XBRL Sponger is the only tool to our knowledge that maps generic

XBRL instance data to RDF [6]. However, in this case, there is not and associated

mapping from the taxonomies instance data is based on to ontology languages.

2 Approach

There are many attempts to move metadata from the XML domain to the Semantic

Web. Some of them just model the XML tree using the RDF primitives [7]. Others

concentrate on modeling the knowledge implicit in XML languages deﬁnitions, i.e.

5 DuCharme, B. Changing my mind about XBRL again, in: Bob DuCharme’s weblog,

http://bobdc.blog, 2008.

6 Raggett, D. XBRL and RDF, in: Dave Raggetts Blog, 2008. http://people.w3.org/ dsr/blog/?p=8

7 XBRL Ontology Speciﬁcation Group, http://groups.google.com/group/xbrl-ontology-

speciﬁcation-group

110

Roberto García and Rosa Gil

DTDs or the XML Schemas, using web ontology languages [8][9]. Finally, there are

attempts to encode XML semantics integrating RDF into XML documents [10][11].

However, none of them facilitate an extensive transfer of XML metadata to the

Semantic Web in a general and transparent way. Their main problem is that the XML

Schema implicit semantics are not made explicit when XML metadata instantiating

this schemas is mapped. This is so because the RDF data produced from XML in-

stance data looses its links to the XML Schemas that structure them and model the

relations among different XML entities.

These relations among different XML entities are what carry the XML Schema

implicit semantics. They capture part of the meaning intended by the schema de-

veloper that, though XML Schema does not provide a way to encode semantics, is

recorded in the way XML Schema constructs are used. For instance, by modeling

that element “father” is a substitutionGroup for element “parent”, it is possible to

interpret that “parent” is more general than “father” and that “father” can appear

where “parent” appears. More details about the implicit semantics of XML Schema

constructs as compared to OWL ones are provided in Section 2.1.

Therefore, the previous mappings from XML to RDF do not take proﬁt from

the meaning encoded in XML Schemas and produce RDF metadata almost as

semantics-blind as the original XML. Or, on the other hand, they capture this seman-

tics but they use additional ad-hoc semantic constructs that produce less transparent

metadata.

Therefore, we have chosen the XML Semantics Reuse methodology [12] and

the XML Schema to OWL and XML to RDF tools implemented in the ReDeFer

project 8 . This methodology combines an XML Schema to web ontology mapping,

called XSD2OWL, with a transparent mapping from XML to RDF, XML2RDF. The

ontologies generated by XSD2OWL are used during the XML to RDF step in order

to generate semantic metadata that takes into account the XML Schema intended

meaning.

This approach has already shown its usefulness with other quite big XML

Schemas in the Digital Rights Management domain, such as MPEG-21 and ODRL

[13], and al-so in the E-Business [14] and multimedia metadata domains [15], where

it produced the more complete MPEG-7 ontology to date [16].

2.1 XSD2OWL Mapping

The XML Schema to OWL mapping is responsible for capturing the schema implicit

semantics, which is determined by the combination of XML Schema constructs. The

mapping is based on translating these constructs to the OWL ones that best capture

their intended meaning. These translations are detailed in Table 1 and Table 2 shows

an example mapping.

8

ReDeFer project, http://rhizomik.net/redefer

Linking XBRL Financial Data

111

The XSD2OWL mapping is quite transparent and captures a great part XML

Schema semantics. The same names used for XML constructs are used for OWL

ones, although in the new namespace deﬁned for the ontology. XSD and OWL con-

structs names are identical; this usually produces uppercase-named OWL properties

because the corresponding element name is uppercase, although this is not the usual

convention in OWL.

Table 1 XSD2OWL translations for the XML Schema constructs

XML Schema OWL Mapping Motivation

element | attribute rdf:Property Named relation between nodes or nodes

owl:DatatypeProperty and values

owl:ObjectProperty

element@substitutionGroup rdfs:subPropertyOf Relation can appear in place of a more gen-

eral one

element@type

rdfs:range

The relation range kind

complexType | group

owl:Class

Relations and contextual restrictions

| attributeGroup

package

complexType / element

owl:Restriction

Contextualised restriction of a relation

extension@base | rdfs:subClassOf

Package concretises the base package

restriction@base

@maxOccurs

owl:maxCardinality Restrict the number of occurrences of a

@minOccurs

owl:minCardinality relation

sequence

owl:intersectionOf

Combination of relations in a context

choice

owl:unionOf

Therefore, XSD2OWL produces OWL ontologies that make explicit the seman-

tics of the corresponding XML Schemas. Table 2 shows a piece of an XML Schema

and the OWL that is generated following this approach.

The only caveats are the implicit order conveyed by xsd:sequence and the ex-

clusivity of xsd:choice . For the ﬁrst problem, owl:intersectionOf does not retain its

operands order, there is no clear solution that retains the great level of transparency

that has been achieved. The use of RDF Lists might impose order but introduces

ad-hoc constructs not present in the original metadata.

Table 2 XSD2OWL translations for the XML Schema constructs

XML Schema

OWL (Abstract Syntax)

< complexType name=“contextOrganisationType” >

Class (contextOrganisationType

< complexContent >

complete

< extension base=“contextEntityType” >

contextEntityType

< sequence >

restriction(Country

< element name=“Country”

allValuesFrom(CountryType)

type=“CountryType”/ >

cardinality(1)))

< /sequence >

< /extension >

< /complexContent >

< /complexType >

112

Roberto García and Rosa Gil

Moreover, as it has been demonstrated in the Semantic Web community, the el-

ement ordering does not contribute much from a semantic and knowledge repre-

sentation point of view [17] in most cases and when it is a requirement it is more

convenient to explicitly represent it using some sort of order attribute or property.

For the second problem, owl:unionOf is an inclusive union, the solution is to use the

disjointness OWL construct, owl:disjointWith , between all union operands in order

to make it exclusive.

2.2 XML2RDF Mapping

Once all the metadata XML Schemas are available as mapped OWL ontologies, it

is time to map the XML metadata that instantiates them. The intention is to pro-

duce RDF metadata as transparently as possible. Therefore, a structure-mapping

approach has been selected [7] instead of a model-mapping one [18].

XML model-mapping is based on representing the XML information set using

semantic tools. This approach is better when XML metadata is semantically ex-

ploited for speciﬁc purposes. However, when the objective is to obtain semantic

metadata from different kinds of input XML data, it is better to follow a more trans-

parent approach.

Transparency is achieved in structure-mapping models because they only try to

represent the XML metadata structure, i.e. a tree, using RDF. The RDF model is

based on the graph so it is easy to model a tree using it. Moreover, we do not need

to worry about the semantics loose produced by structure-mapping. We have for-

malised the underlying semantics into the corresponding ontologies and we will

attach them to RDF metadata using the instantiation relation rdf:type later.

The structure-mapping is based on translating XML metadata instances to

RDF ones that instantiate the corresponding constructs in OWL. The more ba-

sic translation is between relation instances, from xsd:elements and xsd:attributes

to rdf:Properties . Concretely, owl:ObjectProperties for node to node relations and

owl:DatatypeProperties for node to value ones.

Values are kept during the translation as simple types and RDF blank nodes are

introduced in the RDF model in order to serve as the source and destination for

properties. They will remain blank for the moment until they are enriched with

semantic information.

The resulting RDF graph model contains all that we can obtain from the XML

tree. It is already semantically enriched thanks to the rdf:type relation that connects

each RDF property to the owl:ObjectProperty or owl:DatatypeProperty it instanti-

ates. It can be enriched further if the blank nodes are related to the owl:Class that de-

ﬁnes the package of properties and associated restrictions they contain, i.e. the cor-

responding xsd:complexType . This semantic decoration of the graph is formalised

using rdf:type relations from blank nodes to the corresponding OWL classes.

At this point we have obtained a semantically enabled representation of the in-

put metadata, a representation that makes the meaning intended by the XML and

Linking XBRL Financial Data

113

XML Schema modelers explicit from a computer point of view. The instantiation

relations can now be used to apply OWL semantics to metadata. Therefore, the se-

mantics derived from further enrichments of the ontologies, e.g. integration links

between different ontologies or semantic rules, are automatically propagated to in-

stance metadata thanks to inference.

2.3 Algorithm

Listing 2 shows part of the algorithm that implements the XML to RDF mapping.

Basically, starting from the root element, it traverses the XML tree and produces

triples for all attributes and elements recursively using the “mapResProps” method.

All the references to the traversed elements and their attributes are mapped to their

equivalent in the OWL ontologies corresponding to the original XML Schemas.

This is done by the “map” function.

3 Results

First of all, we have generated an ontological infrastructure for the XBRL core,

currently XBRL 2.1. It is composed by the ontologies resulting from mapping the

XBRL core XML Schemas using the XSD2OWL mapping: XBRL Instance, XBRL

Linkbase, XBRL XL and XBRL XLink.

Apart from the previous generic schemas, the schemas listed at the end of this

section have been also mapped in order to be able to map the XBRL data submitted

to the SECs EDGAR program. These schemas are part of the EDGAR Standard Tax-

onomies. The US Financial Reporting - February 28, 2005 taxonomies have been

considered as they are used by the input data currently submitted to this program.

Listing 2 XML2RDF Algorithm

Model XML2RDF( Document d )

{

Model r d f ;

R e s o u r c e r = r d f . c r e a t e R e s o u r c e ( doc . u r l ) ;

E l e m e n t e = doc . g e t D o c u m e n t E l e m e n t ( ) ;

P r o p e r t y p = map ( e . nsURI ( ) ) + e . l o c a l N a m e ( ) ;

C l a s s r a n g e = map . g e t P r o p e r t y R a n g e ( n u l l , p ) ;

r . a d d P r o p e r t y ( RDF . t y p e , r a n g e ) ;

m apR esPro p s ( r , e , r a n g e , r d f ) ;

}

m apR esProp s ( R e s o u r c e r , E l e m e n t e , C l a s s domain , Model r d f )

{

f o r e a c h ( a i n e . a t t r i b u t e s ( ) )

{

P r o p e r t y p = map ( a . nsURI ( ) ) + a . l o c a l N a m e ( ) ;

114

Roberto García and Rosa Gil

r . a d d P r o p e r t y ( p , a . g e t V a l u e ( ) ) ;

}

f o r e a c h ( c i n e . c h i l d N o d e s ( ) )

{

i f ( c . i s T e x t N o d e ( ) ) {

P r o p e r t y p = map ( c . nsURI ( ) ) + c . l o c a l N a m e ( ) ;

r . a d d P r o p e r t y ( p , c . g e t V a l u e ( ) ) ;

}

e l s e {

R e s o u r c e rC = r d f . c r e a t e R e s o u r c e ( ) ;

P r o p e r t y p = map ( c . nsURI ( ) ) + c . l o c a l N a m e ( ) ;

r . a d d P r o p e r t y ( p , rC ) ;

C l a s s r a n g e = map . g e t P r o p e r t y R a n g e ( domain , p ) ;

rC . a d d P r o p e r t y ( RDF . t y p e , r a n g e ) ;

m apR esProp s ( rC , c , r a n g e , r d f ) ;

}

From US GAAP (Generally Accepted Accounting Principles) the schemas, and

corresponding ontologies, are: Primary Terms Elements (USFR-PTE), Primary

Terms Relationships (USFR-PTR), Financial Services Terms Elements (USFR-

FSTE), Financial Services Terms Relationships (USFR-FSTR) and Investment

Management Terms Relationships (USFR-IME). For speciﬁc industries: Bank-

ing and Savings Institutions (US-GAAP-BASI), Commercial and Industrial (US-

GAAP-CI), Insurance (US-GAAP-INS) and Investment Management (US-GAAP-

IM).

There are also some non-GAAP schemas that have been also mapped to OWL

ontologies: Accountants Report (USFR-AR), Management Discussion and Analysis

(USFR-MDA), Management Report (USFR-MR) and SEC Certiﬁcations (USFR-

SECCERT).

Schemas mapped:

• US GAAP ( Accepted Accounting Principles):

– Primary Terms Elements (USFR-PTE)

– Primary Terms Relationships (USFR-PTR)

– Financial Services Terms Elements (USFR-FSTE)

– Financial Services Terms Relationships (USFR-FSTR)

– Investment Management Terms Relationships (USFR-IME)

– Industry:

· Banking and Savings Institutions (US-GAAP-BASI)

· Commercial and Industrial (US-GAAP-CI)

· Insurance (US-GAAP-INS)

· Investment Management (US-GAAP-IM)

• Non-GAAP :

– Accountants Report (USFR-AR)

– Management Discussion and Analysis (USFR-MDA)

Linking XBRL Financial Data

115

– Management Report (USFR-MR)

– SEC Certiﬁcations (USFR-SECCERT)

Each ﬁling for the companies participating in the EDGAR program contains and

XBRL XML ﬁle representing the actual ﬁnancial data and also a speciﬁc XML

Schema extending the XBRL core. This schema provides speciﬁc guides for the cor-

responding ﬁnancial data. Both ﬁles are mapped using XML2RDF and XSD2OWL

respectively.

For instance, for Adobe Systems Inc ﬁling on 2008-07-03, there are the adbe-

20080616.xml ﬁle containing the instance data and the adbe-20080530.xsd schema

for data structures speciﬁc for this ﬁling. They are mapped, respectively, to the RDF

ﬁle for instance data adbe-20080616.rdf and the OWL ontology adbe-20080530.owl

for the schema.

All the previous ontologies are available from the BizOntos Business Ontologies

web page 9 and the semantic data for all the processed ﬁlings can be queried and

browsed from the Semantic XBRL site 10 . Currently, 489 ﬁlings have been processed

from EDGAR. The combination of all these ﬁlings once mapped to RDF amounts

slightly more than 1 million triples, concretely 1,023,929 triples. A triple is the

minimal component of an RDF graph and corresponds to one of its edges connecting

two of its nodes.

Table 4 in the Evaluation section shows the RDF metadata resulting from apply-

ing the XML2RDF mapping to the XBRL context and fact shown in Listing 1. The

RDF metadata references classes and properties from the OWL ontologies result-

ing from mapping the XML Schemas used in the XML instance. This includes the

XBRL schemas and also those speciﬁc for the concrete ﬁling being processes.

For a more general view of the resulting semantic dataset, Figure 1 shows a dia-

gram of the resulting RDF model. At this step, it is possible to take proﬁt from se-

mantic web technologies in order to facilitate connecting the resulting data to other

datasets, but also to improve the interconnectedness of the dataset. Both processes

are detailed in the next subsections.

3.1 Links to External Data

In order to connect the XBRL RDF dataset with other ones in the Web of Linked

Data, the entities in the XBRL model have been analyzed in order to detect those

also described in other datasets. The more prominent ones are companies, a kind

of EntityType present in most EDGAR ﬁlings. XBRL data provides an identiﬁer

for these entities, the Central Index Key (CIK) number. It is a number given to an

individual or company by the U.S. SEC and used to identify the ﬁlings of a company,

person, or entity in several online databases, including EDGAR.

9

BizOntos, http://rhizomik.net/ontologies/bizontos

10

SemanticXBRL, http://rhizomik.net/semanticxbrl

Linking XBRL Financial Data

117

cation process in order to model the ID in the input XBRL. As mentioned before,

from this link speciﬁcation we are able to get just 5 owl:sameAs links between both

datasets.

The next possibility we have explored is to link resources with almost identical

company names. We have used a combination of the Jaro and Q-Gram similarity

measures implemented by Silk. We have been forced to use a quite high threshold for

accepted links because the presence of quite common words in company names, like

“Inc.”, “Corp.”, “Co.”, “Ltd.”, etc., and their many variants makes it very difﬁcult to

get reliable links based on the company name.

After a review of the links generated using the previous approach we have been

able to generate 27 new owl:sameAs relations between the datasets. This is also a

quite scarce amount given that we currently have 543 companies in our dataset. Our

last attempt to date to generate links to DBPedia is to take proﬁt from the fact that

for 398 companies in our dataset we have the ticker.

The obvious approach is to use the dbpprop:ticker property to generate links to

the corresponding DBPedia resources. However, just 4 of them have this property.

Fortunately, we have observer that many DBPedia companies have alternative URI

based on their ticker. In this case, the approach to specify the links has been to ex-

plore the dbpprop:redirect links pointing to DBPedia public companies and strip the

URI in order to get potential tickers. Eg., dbpedia:Microsoft is dbpprop:redirect of

dbpedia:MSFT. Using this approach we have been able to generate 64 owl:sameAs

links to DBPedia.

This continues to be a quite limited amount so we continue to explore other

ways to generate links to dbpedia. Meanwhile, we have also explored other datasets

we can link to. A really interesting candidate is the U.S. Securities and Exchange

Commission Corporate Ownership RDF Data 12 generated by Joshua Tauberer from

SEC and CorpWatch 13 data.

This is a very interesting dataset because it provides information about who is in

the board of many of these companies and also the subsidiary relation among com-

panies. We can use this data in order to generate complex queries that aggregate the

ﬁnancial data we are triplifying from SEC taking into account groups of companies

that hold different kinds of ownership relations, e.g. are all subsidiaries of the same

company or share board members.

In this case it has been easy to generate the links to this dataset because all com-

panies are identiﬁed using their CIK. Not all of them are providing XBRL ﬁlings

so from a total amount of 543 companies in our dataset and 12589 companies in

the ownership dataset, we have obtained 398 links. Table 3 shows a summary of the

number of links to external datasets and the method employed to generate them.

Finally, the other kind of entities that might be connected to external datasets is

units. The easiest kind of entities is currencies because most of the ﬁlings use the

ISO 4217 code in order to identify them. The rest of the units are speciﬁc to the

filings, for instance there is the “shares” or “pure” units that do not have equivalents

12

http://www.rdfabout.com/demo/sec/

13

http://api.corpwatch.org/

118

Roberto García and Rosa Gil

Table 3 Summary of the number of links to external datasets

Linking Method

# Links to DBPedia

# Links to Corporate Ownership Data

SECs identiﬁer CIK

5

398

Company name

27

Company ticker

64

in other datasets. Consequently, we are just linking currencies to their descriptions

in DBPedia.

3.2 Semantic Integration

Apart from the links to other datasets, the EDGAR dataset resulting from the trans-

formation to Semantic Web technologies can be also enriched with internal links.

As it has been mentioned earlier, each XBRL EDGAR ﬁling consists of an XML

in-stance ﬁle accompanied by a XML Schema taxonomy. This taxonomy is speciﬁc

for the ﬁling, it changes from ﬁling to ﬁling. The taxonomy deﬁnes a set of facts

speciﬁc for the ﬁling. New facts are introduced, other used in previous ﬁlings are

removed and some of them suffer minimal modiﬁcations.

For instance, the 2008-07-03 ﬁling from Adobe Systems Inc. refers to the fact

“InvestmentLeaseReceivable” deﬁned in the adbe-20080530.xsd taxonomy while

the 2008-09-16 ﬁling refers to “Investment In LeaseReceivable” deﬁned in the adbe-

20080829.xsd taxonomy. Apart from these slight differences, many facts appearing

in the earlier ﬁling do not appear in the later and the reverse. This happens even if

both ﬁling are of the same kind; in this case they are both “Form 10-Q – Quarterly

report [Sections 13 or 15(d)]” ﬁlings.

These differences among ﬁlings, even when they are of the same type and from

the same company, make it really difﬁcult to integrate them and to perform queries

crossing individual ﬁling boundaries. Consequently, we have taken proﬁt from the

semantic integration tools provided by Semantic Web technologies. The Web Ontol-

ogy Language (OWL) provides a set of primitives that allow stating that two classes,

two properties or two instances are the same. It is also possible to state that some-

thing is a subclass or subproperty of another class or property respectively, that two

classes are disjoint, etc.

These semantic integration statements are then used by inference reasoners,

which are capable of dealing with their semantics while making their implications

totally transparent for the users or applications using them. For instance, it is pos-

sible to state that, continuing with the previous example, “InvestmentLeaseReceiv-

able” and “InvestmentInLeaseReceivable” are equivalent. Consequently, when the

user queries for any of them, the other will be automatically included in the results.

Unfortunately, the process of detecting equivalent or similar concepts and rela-

tions from different ontologies, called ontology alignment, is a very time consuming

one. Moreover, in the case of the EDGAR XBRL ﬁlings, there are a lot of ontolo-

Linking XBRL Financial Data

119

gies to align because, as has been already mentioned, each ﬁling has its own one.

Consequently, automatic or semiautomatic alignment tools are required in order to

get a scalable solution.

Currently, we have just performed some alignments among the ontologies for

the Adobe Systems Inc. ﬁlings. This alignment process has not been integrated into

the whole XBRL to Semantic Web application yet. For instance, we have applied

the alignment implementation provided by the Falcons tool [20] for the two Adobe

Inc. ontologies commented in this section getting a equivalence matching quality

of 0.988 for the “InvestmentLeaseReceivable” and “InvestmentInLeaseReceivable”

properties.

The maximum quality value is “1.0”, which is has been obtained for the 26 prop-

erties with identical names in both ontologies. Overall, more than 70 have been

obtained with a minimum matching quality of “0.741”. The amount of concepts and

properties with the same or very similar labels seems to indicate that it is possible to

achieve a great degree of semantic integration among the ontologies for the ﬁlings

coming from the same companies. We are currently evaluating the quality of the

alignments generated by different tools and for ﬁlings from different companies.

In any case, we can currently take proﬁt from the fact that most facts in the ﬁlings

are not from these ﬁling speciﬁc taxonomies. Most of them come from the standard

XBRL taxonomies. Consequently, we have focused on integrating and cross query-

ing ﬁlings from the point of view of the facts from these standard taxonomies, as

it is shown in the Evaluation section, where the results of the previous process of

moving XBRL data to the semantic space have been put into practice.

4 Evaluation

The XSD2OWL and XML2RDF mappings have been validated in different ways.

First, we have used OWL validators in order to check the logical consistency of the

resulting ontologies. Once all the ontologies were validated, which also includes

checking that all the dependencies among them are met, we proceeded to put them

into practice, together with the semantic metadata generated by the XML2RDF

mapping.

In parallel with our efforts, the ontologies we have generated for XBRL using the

XSD2OWL mapping are being used by OpenLink Software 14 , who has also tested

them independently. These ontologies have been chosen by OpenLink as the onto-

logical framework for their software component responsible for translating XBRL

data to semantic data based on RDF, which they call the XBRL Sponger.

This parallel effort provides us an independent evaluation of the generated on-

tologies, which they have found as appropriate in order to structure the RDF data

they generate from the XBRL ﬁlings. Moreover, they also generate RDF data from

XBRL so we have also evaluated our XML2RDF mapping in comparison to their

14

OpenLink Software, http://www.openlinksw.com

120

Roberto García and Rosa Gil

mapping. As it is shown below, they have implemented their own mapping for this

step thought their and our instance level mappings are based on the same ontologies.

First of all, there is a signiﬁcant difference in the number of triples generated by

the OpenLink XBRL Sponger and XML2RDF. For instance, for the same EDGAR

XBRL ﬁling 15 , the XBRL Sponger produces 900 triples while XML2RDF produces

4739 triples. One possible reason for this difference is that we have followed quite

different approaches relative to how the original XML tree structure is captured in

the RDF graph. However, there is also a signiﬁcant difference in the amount of

instance data captured in the output RDF. While XBRL to RDF captures all the data

in the original XBRL instance, the XBRL Sponger captures just a small part of it in

comparison.

For instance, Table 4 shows in the ﬁrst row a portion of XBRL XML instance

data from the previous ﬁling. This XBRL corresponds to a context and to a fact

that references de previous context. The second row contains the RDF generated

from the previous XBRL XML by the OpenLink Sponger. As it can be shown, the

result is a “sioc:Container” object for the context object that contains just some

of the properties of the original container plus the fact and its value. Some of the

information for the context and most of it for the fact is not captured. Moreover, the

whole structure is ﬂattened.

On the other hand, the third row in Table 4 shows the mapping for the same

XBRL XML as generated by our XML2RDF mapping. As it can be seen, the re-

sult is much move verbose, even more than the original XBRL. However, it does

capture all the original information and keeps the original structure. Even more, the

original XBRL does not explicitly refer to the XML Schema complexTypes deﬁned

in the schemas and used in the instance data. This information is available in the

XML2RDF semantic data and can be used, together with the hierarchical relations

among complex types, when resolving semantic queries against this data.

Apart from instance data, it is also possible to compare the OWL ontologies

generated following the proposed approach to those available from the two other

initiatives introduced in the related work section. It has not been possible to compare

the instance data generated by these initiatives because it not publicly available nor

documented in the corresponding publications or associated documents.

In relation with [5], which focuses on investment funds taxonomies and their

corresponding ontologies, they also perform an automatic mapping from XBRL

taxonomies to OWL ontologies. However, the mapping is not as complete as the

proposed one, especially in relation with cardinalities. The cardinalities in the input

XBRL taxonomies do not seem to be taken into account and thus the output ontolo-

gies deﬁne all properties as FunctionalProperties or cardinalities equal to one.

Finally, comparing with the results reported in [3], they focus on just one taxon-

omy, the IPP-XBRL taxonomy that was promoted by the Spanish Securities Com-

mission (CNMV) then, and just instance data based on this taxonomy can be gener-

ated.

15

Adobe

Systems

Inc.

EDGAR

ﬁling

2008-07-03,

XBRL

ﬁle:

http://www.sec.gov/Archives/edgar/data/796343/000079634308000005/adbe-20080616.xml

Linking XBRL Financial Data

121

Table 4 XBRL XML instance data example (ﬁrst row), OpenLink XBRL Sponger mapping (sec-

ond row) and XML2RDF XBRL mapping (third row) for the previous example

< context id=“AsOf20061201 Consolidated Unaudited” >

< entity >

< identiﬁer scheme=“http://www.sec.gov/CIK” > 796343 < /identiﬁer >

< segment >< adbe:Consolidated / >< /segment >

< /entity >

< period >

< instant > 2006-12-01 < /instant >

< /period >

< scenario >< adbe:Unaudited / >< /scenario >

< /context >

< usfr-pte:CashCashEquivalents decimals=“-3”

contextRef=“AsOf20061201 Consolidated Unaudited”

unitRef=“USD” > 772500000 < /usfr-pte:CashCashEquivalents >

< sioc:Container rdf:about=AsOf20061201 Consolidated Unaudited >

< olsw:identiﬁer > 796343 < /olsw:identiﬁer >

< olsw:scheme rdf:resource=http://www.sec.gov/CIK/ >

< olsw:instant > 2006-12-01 < /olsw:instant >

< olsw:CashCashEquivalents > 772500000 < /olsw:CashCashEquivalents >

< olsw:has space rdf:resource=&adbe796343;adbe-20080616.xml/ >

< /sioc:Container >

< xbrli:context >

< xbrli:contextType rdf:about=“AsOf20061201 Consolidated Unaudited” >

< xbrli:entity >

< xbrli:contextEntityType rdf:about=“&semxbrl;CIK/796343” >

< xbrli:segment >

< xbrli:segmentType >

< adbe20080530:Consolidated rdf:parseType=“Resource”/ >

< /xbrli:segmentType >

< /xbrli:segment >

< /xbrli:contextEntityType >

< /xbrli:entity >

< xbrli:period >

< xbrli:contextPeriodType >

< xbrli:instant > 2006-12-01 < /xbrli:instant >

< /xbrli:contextPeriodType >

< /xbrli:period >

< xbrli:scenario >

< xbrli:contextScenarioType >

< adbe20080530:Unaudited rdf:parseType=“Resource”/ >

< /xbrli:contextScenarioType >

< /xbrli:scenario >

< /xbrli:contextType >

< xbrli:context >

< xbrli:item >

< usfr-pte:CashCashEquivalents >

< rdf:type rdf:resource=“&xbrli;monetaryItemType”/ >

< xbrli:unitRef rdf:resource=“http://dbpedia.org/resource/USD”/ >

< xbrli:decimals > -3 < /xbrli:decimals >

< xbrli:contextRef rdf:resource=“#AsOf20061201 Consolidated Unaudited”/ >

< rdf:value > 772500000 < /rdf:value >

< /usfr-pte:CashCashEquivalents >

< /xbrli:item >

122

Roberto García and Rosa Gil

4.1 Use Case

As a result of how the original XML tree is semantically enriched when it is mapped

to RDF and how different XML trees are interconnected when mapped to RDF

graphs, it is possible to query and traverse the mix of many XBRL ﬁlings in novel

and more productive ways.

All this functionality has been put into practice for the semantic dataset resulting

from mapping the EDGAR XBRL ﬁlings to RDF. The more than 2 million triples

re-sulting from the mapping have been published online using the Rhizomer tool

[21]. Data can be queried, traversed and edited online 16 through a web user interface

for human users. Moreover, through HTTP and content negotiation, Rhizomer also

makes data available for machine consumption and makes it possible to integrate it

into the Web of Linked Data. The overall architecture of this solution is shown in

Figure 2.

For human users, this tool makes it possible to interact with Semantic Web data

by posing semantic queries through dynamic forms or by browsing the RDF graph

interactively. The entry page provides some sample queries that return an HTML

render-ing of the selected parts of the graph, which can be then used as the starting

point for the browsing steps.

This sample queries illustrate how semantic queries take proﬁt from the hierar-

chical relations in the original XML Schemas, i.e. hierarchies of elements and com-

plex types that are translated to property and class hierarchies respectively. More-

over, there is also a query that exploits the fact that some of the Adobe Systems Inc.

ontologies have been integrated and returns data from different ﬁlings for equivalent

facts with different names.

Finally, there are additional views dynamically plugged in depending on the kind

of resource being browsed. Many of them are the same available from Exhibit [22]

(timeline, map, facets,). In addition to visualization plugins, it is also possible to

integrate other kinds of services that manipulate data.

The whole system is built on top of a OpenLink Virtuoso 17 repository that pro-

vides scalability to more than tens of millions of triples and provides RDF Schema

inferencing and support for OWL equivalence constructs.

5 Conclusions and Future Work

As it has been shown, it is possible to map the XML data for XBRL ﬁlings in order

to generate RDF semantic data that keeps all the original information and structure.

This mapping also includes the involved XML Schemas that structure the XML

data. These schemas are mapped to Web ontologies, which make all the seman-

16

SemanticXBRL, http://rhizomik.net/semanticxbrl/

17

OpenLink Virtuoso open-source edition, http://virtuoso.openlinksw.com/wiki/main/

124

Roberto García and Rosa Gil

user interface. The proposed semantic queries illustrate the beneﬁts of the semantic

integration available once XBRL data is translated to semantic data.

However, it is important to note that we do not see our proposal as an alterna-

tive to XBRL. Semantic Web technologies have some limitations that currently do

not make them a clear alternative to XBRL. For instance, OWL does not provide

the primitives to easily model features available in XBRL like the calculation fa-

cilities provided by calculation linkbases. Moreover, the characteristics of the logic

formalisms underlying OWL might not be the more intuitive choice in some XBRL

use scenarios. For instance, a great part of OWL relies on the Open World Assump-

tion and it is based on restrictions instead of on constraints [5].

On the contrary, we see XBRL and the Semantic Web as clearly complementary.

XBRL can be used for business and ﬁnancial data representation and validation,

while its translation to Semantic Web technologies can be the way to make all this

data publicly available enabling cross analysis of this data thanks to semantic inte-

gration and a graph base model.

This vision must be more deeply tested and validated. In order to do that, we are

currently working on integrating ontology alignment tools into the mapping pro-

cess. This way it is going to be possible to extensively put semantic integration into

practice and test the beneﬁts of cross-ﬁlings semantic queries and browsing.

Another future plan is to exploit XBRL semantic data beyond querying and

brows-ing. In this respect, our idea is to take proﬁt from the Rhizomer human-

Semantic Web interaction platform in order to implement additional ways to interact

with this data. For instance, we are currently evaluating an interactive mechanism

for plotting numeric values available through the Parallax interface to Freebase [23].

This would allow performing semantic queries for speciﬁc facts across different ﬁl-

ings and then plotting their values.

Acknowledgements The work described in this chapter has been partially supported by Spanish

Ministry of Science and Innovation through the Open Platform for Multichannel Content Distribu-

tion Management (OMediaDis) research project (TIN2008-06228).

References

1. Lytras, M., García, R. Semantic Web Applications: A framework for industry and business

exploitation - What is needed for the adoption of the Semantic Web from the market and

industry, International Journal of Knowledge and Learning 4(1) (2008) 93-108.

2. Bizer, C., Heath, T., Idehen, K., Berners-Lee, T. Linked data on the web (LDOW2008), in:

Proceeding of the 17th international conference on World Wide Web, ACM, 2008, pp. 1265-

1266.

3. Núñez, S., de Andrs, J., Gayo, J. E., and Ordoez, P. A Semantic Based Collaborative System

for the Interoperability of XBRL Accounting Information, in: Emerging Technologies and

Information Systems for the Knowledge Society. Lecture Notes in Computer Science Vol.

5288, Springer, Berlin, 2008, pp. 593-599.

4. Hoffman, C. Financial Reporting Using XBRL: IFRS and US GAAP Edition. Lulu.com,

2006.

Linking XBRL Financial Data

125

5. Lara, R., Cantador, I., and Castells, P. Semantic Web Technologies for The Financial Do-

main, in: J. Cardoso and M. Lytras (Eds.), The Semantic Web: Real-World Applications from

Industry. Springer, Berlin, 2008, pp. 41-74.

6. Erling, O., Mikhailov, I. RDF Support in the Virtuoso DBMS, in: Pellegrini, T., Auer,

S., Tochtermann, K., and Schaffert, S. (eds.) Networked Knowledge - Networked Media,

Springer, 2009, pp. 7-24.

7. Klein, M.C.A. Interpreting XML Documents via an RDF Schema Ontology, in: Proceedings

of the 13th Int. Workshop on Database and Expert Systems Applications, DEXA02, IEEE

Computer Society, 2002, pp. 889-894.

8. Amann, B., Beeri, C., Fundulaki, I., Scholl, M. Ontology-Based Integration of XML Web Re-

sources, in: Proceedings of the 1st International Semantic Web Conference, ISWC02. Lecture

Notes in Computer Science, Vol. 2342, Springer, Berlin, 2002, pp. 117-131.

9. Cruz, I., Xiao, H., Hsu, F. An Ontology-based Framework for XML Semantic Integration,

in: Proceedings of the 8th Int. Database Engineering and Applications Symposium, IEEE

Computer Society, 2004, pp. 217- 226.

10. Lakshmanan, L., Sadri, F. Interoperability on XML Data. In Proceedings of the 2nd Interna-

tional Semantic Web Conference, ISWC03, Lecture Notes in Computer Science Vol. 2870,

Springer, Berlin, 2003, pp. 146-163.

11. Patel-Schneider, P.F., Simeon, J. The Yin/Yang web: XML syntax and RDF semantics, in:

Proceedings of the 11th World Wide Web Conference, WWW02. ACM Press, 2002, pp. 443-

453.

12. García, R. XML Semantics Reuse, Chapter 7 in: A Semantic Web Approach to Digi-

tal Rights Management, PhD Thesis, Universitat Pompeu Fabra, Barcelona, Spain, 2006.

http://rhizomik.net/ roberto/thesis

13. García, R., Gil, R., Delgado, J. A Web Ontologies Framework for Digital Rights Management,

Journal of Artiﬁcial Intelligence and Law 15, 2 (2007) 137-154.

14. García, R., Gil, R. Facilitating Business Interoperability from the Semantic Web, in: Proceed-

ings of the 10th International Conference on Business Information Systems, BIS’07, Lecture

Notes in Computer Science Vol. 4439, Springer, Berlin, 2007, pp. 220-232.

15. García, R., Perdrix, F., Gil, R., Oliva, M. The Semantic Web as a Newspaper Media Conver-

gence Facilitator, Journal of Web Semantics 6, 2 (2008) 151-161.

16. García, R., Tsinaraki, C., Celma, O., Christodoulakis, S. Multimedia Content Description

using Semantic Web Languages, in: Semantic Multimedia and Ontologies: Theory and Ap-

plications, Y. Kompatsiaris and P. Hobson Eds. Springer, Berlin, 2008, pp. 17-54.

17. Berners-Lee, T. Why RDF model is different from the XML model. W3C Design Issues,

1998. http://www.w3.org/DesignIssues/RDF-XML.html

18. Tous, R., García, R., Rodrguez, E., and Delgado, J. Arquitecture of a Semantic XPath Pro-

cessor, in: Proceedings of 6th Int. Conference on E-Commerce and Web Technologies, K.

Bauknecht, B. Proll and H. Werthner Eds., EC-Web05, Lecture Notes in Computer Science

Vol. 3590, Springer, Berlin, 2005, pp. 1-10.

19. Volz, J., Bizer, C., Gaedke, M., Kobilarov, G. 2009. Silk A Link Discovery Framework for

the Web of Data. 2nd Workshop about Linked Data on the Web (LDOW2009), Madrid, Spain.

20. Hu, W., Qu, Y. Falcon-AO: A practical ontology matching system. Journal of Web Semantics,

6, 3 (2008) 237-239.

21. García, R., Gimeno, J.M., Perdrix, F., Gil, R., Oliva, M. 2008. A Platform for Object-Action

Semantic Web Interaction, in: Proceedings of the 16th Int. Conf. on Knowledge Engineer-

ing and Knowledge Management Patterns, A. Gangemi, J. Euzenat Eds., EKAW08. Lecture

Notes in Computer Science Vol. 5268, Springer, Berlin, pp. 404-418.

22. Huynh, D. User Interfaces Supporting Casual Data-Centric Interactions on

the Web. Doctoral Thesis at MIT EECS / CSAIL, 2007. Available from

http://davidhuynh.net/media/thesis/thesis.php

23. Huynh, D., Karger, D. Parallax and Companion: Set-based Browsing for the

Data Web. Submitted to the World Wide Web Conference, 2009. Available from

http://davidhuynh.net/media/papers/2009/www2009-parallax.pdf