=Paper=
{{Paper
|id=None
|storemode=property
|title=Bringing Parliamentary Debates to the Semantic Web
|pdfUrl=https://ceur-ws.org/Vol-902/paper_6.pdf
|volume=Vol-902
|dblpUrl=https://dblp.org/rec/conf/semweb/JuricHH12
}}
==Bringing Parliamentary Debates to the Semantic Web==
https://ceur-ws.org/Vol-902/paper_6.pdf
Bringing parliamentary debates to the Semantic Web
Damir Juric1,3, Laura Hollink2, Geert-Jan Houben1
1
Delft University of Technology, 2 VU University Amsterdam
3
FER University of Zagreb
Abstract. An analysis of parliamentary debates and media resources that cover
them can provide insight into the political climate of a country. Although
debates are now regularly published on official government portals, their
analysis remains a cumbersome and challenging task for historians and political
scientists. One of the main tasks of the PoliMedia project is to allow easy cross-
media comparisons and give better insight into choices that different types of
media outlets make when covering parliamentary debates. As a first step of that
task, Dutch parliamentary debate data available in XML files is being translated
into Semantic Web standards, which will allow users to easily query the data. In
this paper we discuss design choices we made to build a semantic model that
will represent events and topics from the Dutch parliamentary debates.
Keywords: XML, RDF, SEM, Semantic Web, parliamentary debates
1 Introduction
In this paper we discuss ongoing work on the representation of political events on
the Semantic Web. We present the design choices of a model in which we capture
parliamentary debates, including how they are covered by various media.
Analyzing media coverage across several types of media outlets is a challenging
task, especially for people who need deep understanding of the data and its
implications, like media historians. Previous research has focused mainly on
newspaper articles, because they are generally available in digital, computer-readable
format. To make cross-media comparisons between different types of media outlets,
links between datasets would need to be produced. For example, to support
researchers that want to know how political debates are represented in the media and
how the representation of topics and people change over time. We aim to facilitate
this kind of analysis by providing links between datasets of political debate events and
media data.
The media-historic research questions that guide the project are: “What choices do
different media make in the coverage of people and topics while reporting on debates
in the Dutch parliament since the first televised evening news in 1956 until 1995?
Does the representation of topics and people change over time and how do the
various media types differ?” These questions specify a number of things that needs to
be expressed in the model, like people, topics, time and media types.
To answer these questions we first created a semantic model that is expressive
enough to allow us to represent all important information about events from the Dutch
parliament, that are recorded in the form of debate transcripts (and later in XML
files). After this step, an RDF repository is created in which we instantiate the model
with instances of debate events, that allows various interesting information to be
extracted from this dataset using SPARQL queries.
This paper is organized as follows: first we describe the PoliMedia project in
which this work is carried out. In Section 2 we give a description of our datasets of
debate events and media items. In Section 3 we discus the semantic model, and in
Section 4 we describe our future work.
1.1 Background: the PoliMedia project
The PoliMedia project1 is driven by research questions from historians with respect
to media coverage across several types of media outlets. Cross-media comparisons
will be conducted over a longer period of time, on different topics. The project will
focus on the coverage of the debates in the Dutch parliament and give insight on the
different choices that different media make while reporting on those debates. Also,
when research can be performed with time and media type in mind, another question
can be answered: Does the representation of topics and people change over time and
how do the various media types differ?
The project will be carried out in three phases: (1) a modeling phase: creating a
semantic model (that is the phase described in this paper), (2) a data production phase:
creating links between debates and associated media sources, and (3) an application
phase: searching and navigating linked datasets.
1.2 Related work
Related work for this project comes from three domains: other projects using
parliamentary debate data, event modeling and relatedness discovery.
In [1], the author describes the structure of parliamentary proceedings and sketches
a widely applicable DTD. He also describes how proceedings in PDF format can be
transformed into deeply nested XML files. The work described is done as part of a
project called War In Parliament [2]. In the work described in this paper, we use
structured XML files from War In Parliament as a basis for our model. This dataset
can be searched on the Political Mashup portal [3]. [4] presents an approach that
extends existing metadata enrichment processes with a method to discover historical
events. The events are structured in a historical event thesaurus to enrich object
metadata. As such, the event thesaurus is used as a bridge between objects in different
collections. The results of the approach allows for topic-based and event-centered
browsing, searching and navigating in integrated collections. In [5], the authors put
events as the central elements in the representation of data from domains such as
history, cultural heritage, multimedia and geography. The Simple Event Model (SEM)
1 PoliMedia project: http://www.polimedia.nl/
2
is created to model events in these various domains, without making assumptions
about the domain-specific vocabularies used. The researchers designed SEM with a
minimum of semantic commitment to guarantee maximal interoperability. In [6] the
authors describe real life problem using SEM. Some properties of SEM are used in
the semantic model described in this paper. We used SEM model as a starting point
on which we build our own model that conforms to the events in the parliament. The
problem of link discovery is tackled in [7]: there a validation approach is presented of
detected alignment links between dialog transcript and discussed documents, in the
context of a multimodal document alignment framework of multimedia events
(meetings and lectures). The validation approach consists of an entailment process of
the detected alignment links. This entailment process exploits several features, from
the structural level of aligned documents to the linguistic level of their tokens. In [8]
the authors present a function that discovers relatedness between news articles across
four aspects: relevance, novelty, connection clarity, and transition smoothness.
2 Description of datasets
The PoliMedia project is aimed at cross-linking four different datasets, each from
different media outlets. All datasets, which are textual and audiovisual, are available
via the CLARIN infrastructure.
Primary dataset for this project is a collection of Dutch parliamentary debates, the
so-called Handelingen der Staten-Generaal or the Dutch Hansard. Parliamentary
debates used in this project, are actual transcripts of speeches that politicians had in
the parliament. At the time of writing this article, three sources of Dutch
parliamentary debates are available online. On the Officiële Bekendmakingen portal,
which is an official source for parliamentary debates from the Dutch parliament,
debates can be found in an XML format, using XML schema and permanent
identifiers. Existing identifiers point only to the debate as a whole, not specifically to
parts of the debate. Also, only debates from 1995 till present are available at this
source.
A second source for Dutch parliamentary debates can be found online, on the
Staten-Generaal Digitaal portal2, which contains debates from the parliament from
before the year 1995. Data can be accessed publically using the SRU (Search and
Retrieval via URL) [9] or OAI (Open Archives Initiative) [10] protocols. Contrary to
the previous source, debates from this source posses no further structure (data is
provided in txt or pdf formats).
A third source for political debates from the Dutch parliament can be found on
Political Mashup [3]. This data is created by the CLARIN project War in Parliament
(WIP). The project is still ongoing, and the way debates are published is continuously
improving. At the time of writing this article, all debates until the year 1995 are
published as XML documents (OCR with satisfactory quality is being used). This data
shows a fine-grained structure.
2 Staten-Generaal Digitaal: http://www.statengeneraaldigitaal.nl/
3
Secondary datasets contains different media types: newspaper articles, radio
bulletins, and newscasts. The newspaper and radio bulletins dataset is available from
the National Library of the Netherlands, which allows users to analyze the text of the
articles and see in which way they are layered. Metadata of the articles and bulletins
are available from the metadata store of the Koninklijke Bibliotheek (KB), the KB-
MDO or Koninklijke Bibliotheek metadataopslag [11] as DIDL (Digital Item
Declaration Language – an XML dialect [12]). The newscast dataset contains evening
news and current affairs programs. Audiovisual content include program level
metadata in Dublin Core and CDMI format, enriched with thesaurus terms from the
Gemeenschappelijke Thesaurus Audiovisuele Archieven (GTAA). Data can be
accessed using the OAI-PMH protocol.
3 Semantic model
The semantic model for the PoliMedia project is built to satisfy the requirements of
the project, i.e. the research questions from the users. The model is based on the
Simple Event Model [5] developed in the NWO CATCH project Agora. SEM is a
model to represent events on the Web, and to explicate complicated semantic relations
between people, places, actions and objects: not only who did what, when and where,
but also the roles each actor played, the time during which this role is valid and the
authority according to whom this role is assigned. Because the PoliMedia project
deals with a specific domain, our semantic model is adapted to it so it can express
important information associated with the events and actors in political debates.
3.1 Requirements for the semantic model
The goal of the project is to publish the links on the Web, so using open Web formats
and standards, a Web query language, and unique identifiers (URI’s) is compulsory.
The semantic model of the PoliMedia project is to be expressive in a way that it
allows important information regarding parliamentary debates to be easily accessed.
Important information for every parliamentary debate is:
• The time on which the debate is held
• What is being said in the debate (topics)
• Who is giving the speeches in the debate and in which role (persons)
• Links to additional information about actors involved in the event (names
of the politicians, their party, age, etc.)
• Subparts of the debate have their own identifiers (part of the debate where
only one speaker can be identified as actor)
• Important information about subparts is their chronological order (the order
in which the subparts where occurring inside the parliament debate,
• Named entities apart from politicians (persons, locations, etc.)
Important information for parliamentary debates that are specific to PoliMedia
project:
4
• Links between subparts of the debate and news articles, radio bulletins and
television newscasts
• Various information about media items linked to the debate
Data from the parliamentary debates is available online, so unique identifiers are
created for:
• Debates (as a document as a whole) and for the parts of the debates
• Individual news articles, radio bulletins, and television newscasts
• All political parties of the speakers in the debates as well as the speakers
them self
All important information about debates listed here are represented in the semantic
model.
3.2 URIs as identifiers
On the Semantic Web, all entities are identified by a URI. In our case, all source
datasets already contain URIs. Our preference is to use these existing URIs directly
instead of creating our own URIs. For example, we link to the newspapers of the
Koninklijke Bibliotheek with statements like:
"http://resolver.kb.nl/resolve?urn=ddd:010688440:mpeg21:a0001:ocr" ;
We have made a different choice for the debate events, as these are the core of our
dataset. Also for debates, URIs do already exist: the government website
officielebekendmakingen.nl provides persistent URIs to debates after 1995, and the
project War In Parliament provides URIs for debates as well as parts of debates.
Nevertheless, we create our own URIs for each debate and parts of debates, for
example:
a ;
The reason for this is that we want the URIs to be dereferenceable, i.e. we want to
serve informative and descriptive RDF when the event URI is requested. Neither
officielebekendmakingen.nl nor War in Parliament does this. We use so-called
PURLs (Persistent Uniform Resource Locators), Web addresses that act as permanent
identifiers.
3.3 Provenance
We build on existing data and tools. It is important to preserve this provenance
information, both to give credit where credit is due and to provide information about
how much the data can be trusted. For every debate in our model we add information
about the original source of the debate. For example, the next statement uses the
dc:source property to state that the original debate came from Political Mashup:
"http://resolver.politicalmashup.nl/nl.proc.sgd.d.19590000048" ;
5
Named entities were extracted in the War In Parliament project using the NER tool
Folia[14]. We use the dc:provenance property to state the source of extracted entity.
3.4 Description of the semantic model
The semantic model, as well as the links between datasets, is expressed in the RDF
format, W3C Standard for Semantic Web. Also, the data is made compatible with the
ISOCAT standard3, Dublin Core4 and SKOS5.
We created this semantic model to conform to the rules and regulations of the
Dutch parliament, although the model can be easily adapted to follow different rules
(of parliaments in other countries), because in its core all parliamentary debates
consists of the same most important elements like the topics and the speeches.
All debates conform to the same rule, where speakers give speeches in the
parliament in some chronological order. First speaker is always the “voorzitter” (the
person who is in charge of the actual debate and can be called chairman). The
chairman gives usually an introduction to the topic and after his speech he gives the
floor to some member of the parliament.
Every debate has three main structural elements:
• The topics – the themes or agenda of the meeting
• The speeches – every word by every speaker is transcribed including the
names of the speakers and their affiliation
• Actions – descriptions, lists, etc.
Every transcript contains metadata with important information about the debate as
a whole, like the date when the debate actually happened in the parliament, the title of
the debate etc. In the PoliMedia semantic model, as can be seen in Fig1., a debate is
represented as a resource with its own unique identifier (for example:
http://purl.org/linkedpolitics/nl/nl.proc.sgd.d.19720000002). This resource serves as a
domain for Dublin Core properties like dc:date, dc:title, dc:identifier, dc:publisher,
dc:source and dc:language, which points to the literals that contain information about
the date when the debate happened, its official title, unique identifier and original
source, the publisher and the language on which the debate was published (an RDF
example is given in Fig. 2).
The PoliMedia specific property hasPart is attached to the resource containing the
debate URI and points to the range of possible parts of the debate that the debate as a
whole can contain (this element is shown in Fig.3). One specific part of a debate
always contains elements called DebateContext and Speech. Element DebateContext
contains text that is read by the chairman (voorzitter) of the debate and that text
represents the short description of subjects that will be addressed in the forthcoming
speech.
3 http://www.isocat.org/
4 http://dublincore.org/
5 http://www.w3.org/2004/02/skos/
6
Fig. 1. Part of semantic model representation of the debates dataset (second part on
Fig. 4.)
Fig. 2. Debate represented in RD
7
Fig. 3. Part of the debate represented in RDF
The most important element of the PoliMedia semantic model is the element Speech
that represents the actual speech that a certain member of Parliament has spoken
while addressing the issues of the debate topic (Fig.4 and Fig.5). The content of the
speech is saved as a Literal. Every speech has its speaker and those two resources are
connected with the sem:hasActor property described in the Simple Event Model[5].
Property hasActor points to the blank node with three other properties leaving from
the node. Objects of those properties are URIs that lead to the pages of the politician
giving the speech, to the party the mentioned politician is member of, and SEM
properties denoting the role of the hasActor property.
Fig. 4. Semantic model representation of the debates dataset
8
Fig. 5. Example of one speech in RDF
By nature, speeches in the parliament usually contain a great number of named
entities, such as names of politicians or business people, names of different
organizations, and geographical locations. Named entities were recognized in
parliamentary debates in the project War in Parliament. Names of persons,
organizations, locations, and miscellaneous entities were extracted from transcripts
using a tool for Linguistic Annotation[13]. Named entities are connected with four
different properties where each one points to different objects of the triple (either
person, location, organization or miscellaneous entity). A literal is created for every
named entity found in the speech together with a URI that leads to the Wikipedia page
of the entity, in case that page exists.
The semantic model for secondary datasets is straightforward. Both SRU and OAI-
PMH protocols allow the client to submit a search and retrieve request for matching
records from the secondary datasets. A response on a query containing the matched
keywords contains Dublin core properties such as dc:identifier, dc:type, dc:publisher,
dc:date, dc:source and dc:title which are used in our PoliMedia semantic model in
case of newspaper articles. The model will contain the instance of a newspaper article
with a URI that uses a resolver for accessing the OCR text or pdf document at the
National Library. Both radio bulletins and newscast datasets have very similar
models. The newscast dataset contains very rich metadata about its resources, so
except information about the date, type and publisher, this metadata contains spatial
information and names of subject that appears in the videos.
As a final result of the first phase of our project, we created an RDF repository that
contains around 38,8 million triples, that came from 10,924 XML files containing
information about debates in Dutch parliament. Important elements from XML files
were extracted using Java libraries (SAX) and RDF triples were created (JENA). The
semantic repository is created using OWLIM6, a software component for storing and
manipulating huge quantities of RDF data. OWLIM is packaged as a Storage and
Inference Layer (SAIL) for the Sesame OpenRDF framework.
6 OWLIM – Semantic repository: http://owlim.ontotext.com/display/OWLIMv51/Home
9
4 Summary and next steps
In this paper we described the process of creating the semantic model for the purpose
of building a semantic repository for the PoliMedia project. The semantic repository
is filled with triples that describe events and topics that happened in the Dutch
parliament and allows us to use queries to fetch interesting information that was not
as easily available before (for example, how many times a particular politician spoke
of a particular person in the parliament).
As previously stated, the PoliMedia project will be carried out in three phases. Phases
that will be carried in the future are phases (2) and (3), with an automatic detection of
the semantic links between primary and secondary datasets and the creation of a
demonstrator application.
For the creation of links Named Entities (that appears in primary and secondary
datasets) will be used to decide whether the media resource is on some way connected
to the events discussed in the debates. Important entities are persons but also locations
and time. As debate events consist of smaller sub-events, namely speeches of
consecutive speakers (as it is expressed in the semantic model described in this
paper), we will search for possible links between those sub-events and media items
that cover that particular part of the debate. A virtual research environment will be
built that allows the exploration of the debate events and media coverage thereof via
search and browsing. Next to the use of standard information retrieval libraries
(Lucene), navigation options will be implemented that will allow users to browse
through the linked datasets of debates and media.
References
1. Maarten Marx: Advanced Information Access to Parliamentary Debates. J. Digit. Inf. 10(6):
(2009)
2. War In Parliament project: http://www.clarin.nl/page/about/projects/162#WIP
3. Political Mashup project: http://politicalmashup.nl/
4. van Erp, Marieke et al, :Automatic Heritage Metadata Enrichment with Historic Events. In
J. Trant and D. Bearman (eds). Museums and the Web 2011: Proceedings. Toronto:
Archives & Museum Informatics. Published March 31, 2011. Consulted March 5, (2012).
5. W. van Hage, V. Malaisé, R. Segers, L. Hollink, and G. Schreiber: Design and use of the
Simple Event Model (SEM). J. Web Semantics, 2011.
6. Hage, W.R. van, V. Malaisé, G. de Vries, G. Schreiber and M. van Someren: “Combining
Ship Trajectories and Semantics with the Simple Event Model (SEM)”. In: Proceedings of
the 1 stACM International Workshop on Events in Multimedia 73-80, (2009)
7. Dalila Mekhaldi, Denis Lalanne: Multimodal Document Alignment: Feature-based
Validation to Strengthen Thematic Links. JMPT 1(1): 30-46, (2010)
8. Y. Lv, T. Moon, P. Kolari, Z. Zheng, X. Wang, and Y. Chang: Learning to model
relatedness for news recommendation. In WWW, (2011).
9. SRU: Search/Retrieval via URL: http://www.loc.gov/standards/sru/
10. OAI protocol: http://www.openarchives.org/OAI/openarchivesprotocol.html
11. Koninklijke Bibliotheek metadataopslag:http://research.kb.nl/documenten.html
12. Digital Item Declaration Language: http://xml.coverpages.org/mpeg21-didl.html
13. FoLiA: Format for Linguistic Annotation: http://ilk.uvt.nl/folia/
10