=Paper=
{{Paper
|id=Vol-1529/paper7
|storemode=property
|title=Entity-Centric Preservation for Linked Open Data: Use Cases, Requirements and Models
|pdfUrl=https://ceur-ws.org/Vol-1529/paper7.pdf
|volume=Vol-1529
|authors=Elena Demidova,Thomas Risse,Giang Binh Tran,Gerhard Gossen
|dblpUrl=https://dblp.org/rec/conf/ercimdl/DemidovaRTG15
}}
==Entity-Centric Preservation for Linked Open Data: Use Cases, Requirements and Models==
https://ceur-ws.org/Vol-1529/paper7.pdf
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
Entity-Centric Preservation for Linked Open Data: Use
Cases, Requirements and Models
Elena Demidova, Thomas Risse, Giang Binh Tran, and Gerhard Gossen
L3S Research Center & Leibniz University Hannover
Appelstrasse 9, 30167 Hannover, Germany
{demidova,risse,gtran,gossen}@L3S.DE
Abstract. Linked Open Data (LOD) plays an increasingly important role in the
area of digital libraries and archives. It enables better semantic-based access
methods and also gives the human reader more insights into linked entities. Since
semantics is evolving over time, the temporal dimension has an increasing im-
pact on the applicability of LOD to long-term archives. LOD does not necessarily
provide a history. Therefore it is not possible to go back in time and access the
document related LOD content at the time of the document creation. In this pa-
per we propose to collect LOD information along with the Web content for Web
archives to also ensure a good semantic coverage of a Web crawl. We discuss use
cases and requirements and derive the general approach and related data models
for the implementation.
Keywords: Linked Open Data, Entity Preservation
1 Introduction and Motivation
Linked Open Data (LOD) plays an increasingly important role in the area of digital
libraries and archives. LOD is a standardized method to publish and interlink structured
semantic data. Major LOD hubs like DBpedia1 or Freebase2 extract information from
Wikipedia and provide it as structured data with well-defined semantics. By linking en-
tities in documents with their semantic descriptions in the LOD Cloud (i.e. the datasets
published in LOD format), richer semantic descriptions of the document itself can be
achieved. On the one hand, this enables better semantic based access method. On the
other hand, since the LOD information is often based on Wikipedia and other textual
entity descriptions, it also gives the human reader more insights into the entity at crawl
time.
Although long-term archives can benefit from LOD, the temporal dimension has
an increasing impact on the access and the interpretation of content. The human un-
derstanding of an information object at a later point in time requires, depending on the
kind of information object, additional context information. While traditional materials
like papers or books often bring enough context information for the understanding, this
1
dbpedia.org
2
www.freebase.com
61
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
2
is rarely the case for user generated content on the Web and Social Web. A twitter mes-
sage like “The new #ipod is cool http://bit.ly/1cd7ylZ” will be hardly understandable in
50 years without additional knowledge. We do not know today if iPods will still exist
or if people will know what an iPod was in 2014. Since entities are also evolving (e.g.
names change), we do not even know if the iPod in 2064 is still a portable music player.
While LOD can successfully support the semantic access to content today[9], it
is of limited use for long-term archives. LOD is constructed similar to the Web as a
constantly evolving source of knowledge. In this way it also inherits the property of the
Web of not having a documented history. For the Web this issue is partly addressed by
establishing Web archives or application specific versioning support e.g. in Wikipedia.
For LOD, independent steps in this direction are done by some sites like Freebase to
offer snapshots of their content at different times. However, this is site specific and no
general support exists.
To address these issues we propose, in the area of Web archiving, to preserve rele-
vant parts of the LOD Cloud along with crawled Web pages. This requires a semantic
analysis including entity extraction from Web pages (e.g. using NER techniques) cou-
pled with enrichment of extracted entities using metadata from LOD as close as possible
to the content collection time point and calls for integrated crawling and Web archive
metadata enrichment approaches that jointly collect Web data and related LOD enti-
ties. Since the LOD graph is - like the Web - an ever growing information collection, a
number of issues need to be addressed that we will discuss in this paper. These issues in-
clude integrated crawling approaches to seamlessly collect Web documents along with
relevant entities from the LOD cloud, prioritization methods to select the most relevant
sources, entities and properties for archiving as well as modelling of the entities and
their provenance (in particular the LOD sources) within Web archives. In summary, the
contributions of this paper are as follows:
– Analysis of major use cases to derive requirements for entity-centric LOD preser-
vation in the context of Web archiving;
– An approach for the entity-centric crawling of Linked Open Data content;
– Data models for the entity preservation in Web archives.
This paper is an extended version of [13] that discussed a preliminary approach for
entity-centric crawling of Linked Open Data content.
2 Use Cases
Today an increasing interest in using Web content can be observed within the scientific
community. Long term Web archives are becoming more interesting for researchers in
humanities and social sciences. For example, historical sciences are increasingly using
Web archives for their research. However, major research activities on events and topics
from historian point of view will begin in around 25 years after the event. Due to the
volume of information available on the Web and partly preserved in Web archives,
working with the content raises a number of challenges. Especially temporal aspects
depending on the time span between crawling and usage have an increasing impact on
the way how to find documents and to interpret the content.
62
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
3
Searching for documents on today’s Web is mostly based on specifying the search
intent by using a number of keywords. Web search engines, for example, take a number
of assumptions to fulfill the users’ informational needs. This way, Web search for the
entity “Pope” will result in many pages mentioning Pope Francis as the search engine
assumes that in 2014 most interest exists on the current Pope. This assumption does
not hold for Web archives usage (and other long term archives) containing snapshots
of parts of the Web of many years. Making assumptions about the potentially most
interesting interpretation of a keyword query “Pope” in a Web archive is hard since the
temporal search intentions of the users can be manifold.
Knowledge about temporal aspects of entities can help to disambiguate them and
support the Web archive users. The required knowledge can be taken from LOD sources
like Freebase or DBpedia. By linking entities contained in the archived documents to
their descriptions in the LOD Cloud at the crawling time, the entities get a unique tem-
poral interpretation. For example, the entity “Pope” mentioned in a newspaper article
created in 2012 will most probably linked to Pope Benedict XVI, while in a document
created in 2014 would be linked to Pope Francis. The interlinking helps finding other
documents where the same entity has been mentioned as well as to disambiguate men-
tions of other entities with the same name at a different point in time.
By taking the long-term perspective the temporal dimension gains a crucial impor-
tance. One example is the evolution of entities, and in particular their name changes.
These changes can often be observed for geographic names and organizations. For ex-
ample, the dissolution of the USSR in 1991 resulted in renaming of entire states, cities
and streets throughout the involved countries, often returning locations their historical
names. A prominent example in this context is the change of the name of St. Petersburg
(1703 - 1914 and from 1991) to Petrograd (1914 - 1924) and Leningrad (1924 - 1991).
Also persons are changing names e.g. Pope Francis was named Jorge Mario Bergoglio
before his election in 2013. This is an important information for time-aware query re-
formulation. In this way a query about Pope Francis would be extended with his birth
name to obtain documents written before 2013.
The interlinking of the archived documents with LOD entities can also facilitate
access to human readable presentation of entities on the Web to get insights into the
perception of the entity at the crawling time, which can be especially beneficial for re-
searchers in historical and social sciences. For example, the former Managing Director
of the International Monetary Fund Dominique Strauss-Kahn and the former President
of Germany Christian Wulff have been highly distinguished people before a number of
allegations came up. For getting a better understanding of an archived document in its
temporal context, the reader should know how the entities contained in this document
were described at the creation time or, at latest, at the crawling time.
A time sensitive access to Linked Open Data is currently not supported. Even though
Wikipedia provides a detail history for each document this is not the case for most LOD
sites. This has some consequences for the access and interpretation of entities in the fu-
ture. For example, the LOD description of Pope Benedict XVI in a document created
2012 will present his description of today. Even if the description is correct, it does not
show the perception of the Pope Benedict in 2012. Wikipedia - the source of DBpedia,
Freebase and other - is constantly updated and even if it aims at neutral descriptions,
63
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
4
the perception of the authors has always an impact. For example, an event an entity is
involved in could be highlighted for some time and later be lowered again. The differ-
ences in the descriptions of an entity at different times provide many interesting insights
for researchers. Even the non-existence of a LOD entry in a specific dataset can be of
interest and should be documented.
To enable future researchers to get good and comprehensive overview about topics
and events today, adequate Web archiving including entity-centric LOD preservation is
a necessity. This will ease the access to archived content and support its interpretation
in the future.
3 Requirements
In order to facilitate interpretation of the archived Web documents and LOD entities
linked to these documents in the future, presentation of the entities within Web archives
should take into account content, provenance, quality, authenticity and context dimen-
sions. In the following we present these requirements in more detail.
3.1 Content and Schema
The entities on the Web of Data are typically represented using triples, in the form of
. Thereby entity URI is a unique identifier of the entity
within the data source. The properties can be further differentiated in the datatype prop-
erties (directly providing property values) and object properties (linking the entity to
other entities and facts in the knowledge graph using URIs). Object properties are also
used to link entities to entity types within a classification scheme.
In order to get a complete information provided by the object properties, traversal of
the knowledge graph is required. However, traversal of large knowledge graphs is com-
putationally expensive. Apart of that, while available properties are source dependent
their usefulness with respect to the specific Web archive varies. Whereas some prop-
erties (e.g. entity types, or equivalence links) can be considered of crucial importance,
others can be less important such that they can be excluded from the entity preservation
process. Therefore, property weighting is required to guide an entity-centric crawling
process.
3.2 Provenance and Quality
Similar to Web document archiving, documentation of the crawling or entity extraction
process as well as unique identification and verification methods for collected entities
are required to ensure re-usability and citeability of the collected resources. In addi-
tion, for better interpretation of the entities stored within the Web archive, it is crucial
to collect metadata describing the data source of entity origin. Optimally, these meta-
data should include data quality parameters such as methods used for dataset creation
(e.g. automatic entity extraction, manual population, etc.), freshness (last update at the
dataset and entity levels), data size, as well as completeness and consistency of data.
64
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
5
Unfortunately, such metadata is rarely available in the LOD cloud. Therefore, archiv-
ing systems should provide functionality for statistical analysis of data sources to esti-
mate their quality and reliability. To ensure correct access to the archived information
available metadata about publisher, copyright and licenses of the sources needs to be
preserved.
3.3 Authenticity and Context
Authenticity is the ability to see an entity in the same way as it was present on the
Web at the crawl or extraction time. One of the main principles of LOD is the use of
dereferenceable URIs that can be used to obtain a Web-based view on an entity (this
view also may differ from the machine-readable representation). For example, Figure
1 presents a part of the Web and RDF representations of entity “London” in Freebase
dataset from March, 4, 2014.
(a) Web representation (b) RDF representation
Fig. 1. “London” in Freebase
To satisfy authenticity requirements, in addition to machine-readable entity rep-
resentation, human-readable pages representing archived entities should be preserved.
Such information can include visual resources such as photographs of people, maps
snippets, organization logos, snippets of Wikipedia articles, etc. The archiving of human-
readable entity representation can enable correct entity disambiguation going beyond
the automatic processing and make archived entities valuable assets e.g. for social sci-
ence researchers.
Authenticity raises a number of issues in the dynamic Web environment. One im-
portant issue for the authenticity is that the archived Web pages containing the entity,
the Web pages representing the entity in LOD as well as machine-readable entity repre-
sentation have to be crawled within a short time interval to collect coherent entity views
in both, machine-readable and human-readable representations.
4 Entity Extraction and Preservation
In the context of Web archives, entities can be extracted from the archived Web doc-
uments using Named Entity Recognition (NER) tools as a part of Entity Extraction or
65
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
6
added to the archive from external LOD sources during the Entity Preservation as a
part of overall Web archive metadata enrichment process. Thereby, the nature of LOD,
being a distributed collection of datasets, brings alone several challenges with respect
to entity crawling and preservation.
4.1 Entity Extraction
Entity extraction within Web archives can typically be handled using state-of-the-art
NER tools such as Stanford NER [12], Illinois Named Entity Tagger [24], SENNA [6],
SpeedRead [1], Wikipedia Miner [19], TagMe [11], Illinois Wikifier [5] and others.
These tools make use of different approaches to entity extraction and interlinking and
indicate significant differences with respect to precision, coverage and efficiency.
The Wikipedia-based NER tools such as Wikipedia Miner and Illinois Wikifier di-
rectly support linking of extracted entities to Wikipedia. However, they have a relatively
small coverage being restricted to entities mentioned in Wikipedia and are less efficient
as they require access to Wikipedia dumps during the extraction phase. For example,
Wikipedia Miner – one of the best tools for Wikipedia-based entity recognition – con-
sumes a large chunk of memory to load the Wikipedia entity graph during the process-
ing [1, 7]. Experimental results in the literature also show that text-based tools such
as SENNA, Stanford NER and SpeedRead can achieve better precision and recall than
Wikipedia-based tools.
4.2 Entity Preservation
Whereas entity extraction can deliver entity labels, types and eventually initial pointers
(URIs) of the relevant entities in the reference LOD datasets, the collection of relevant
entities should be extended beyond the LOD sources used by the extractors. Further-
more, the content of the entities needs to be collected and preserved according to the re-
quirements presented in Section 3. In these context important challenges are connected
to SPARQL endpoint discovery and metadata crawling, prioritization of the crawler and
entity-centric crawling to obtain the most relevant parts of the LOD graphs efficiently.
SPARQL endpoint discovery and metadata crawling: Existing dataset catalogs such
as DataHub3 or the LinkedUp catalogue4 include endpoint URLs of selected datasets
as well as selected statistics, mostly concerning the size of specific datasets; however,
existing catalogs are highly incomplete. Furthermore, existing approaches to distributed
SPARQL querying suffer from efficiency, scalability and as a result low availability of
SPARQL endpoints [27]. As an alternative, some LOD sources provide data dumps
for download; however, given the dynamic nature of LOD, the dumps require frequent
updates and come alone with lack of scalable tools for their processing. Approaches to
alleviate these problems are being actively developed in the Semantic Web community
[27].
3
http://datahub.io/
4
http://data.linkededucation.org/linkedup/catalog/
66
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
7
Metadata crawling includes several steps to obtain SPARQL endpoints URLs and
generate corresponding metadata. Based on this metadata, prioritization of LOD sources
and properties within these sources for preservation can be performed. Figure 2(a)
presents the overall procedure of SPARQL endpoint discovery and metadata crawling.
In total, SPARQL endpoint discovery, metadata collection and pre-processing include
following steps:
(a) SPARQL endpoint discovery and metadata (b) Entity-centric crawling
crawling
Fig. 2. Entity preservation
Step 1a. Query LOD catalogues to obtain a seed list of LOD datasets, SPARQL end-
points and other available metadata. The crawler can also work with a pre-defined
seed list of LOD sources.
Step 1b. For each available LOD source, collect and generate metadata. The metadata
should include topical information to facilitate selection of relevant datasets for a
specific Web archive, available schema information (especially with respect to the
relevant entity types and their properties), the version of the dataset, quality-related
statistics and license information.
Step 1c. For a specific Web archive, perform LOD dataset selection according to the
topical relevance and quality parameters. If schema information is available, estab-
lish property weighting for crawl prioritization.
Prioritization of the crawler: Similar to Web crawling, there is a trade-off between
data collection efficiency and completeness in the context of LOD crawling. On the one
hand, the entity preservation process should aim to create a possibly complete overview
of the entity representations available in LOD and collect information from possibly
many sources. On the other hand, due to the topical variety, scale and quality differ-
ences of LOD datasets, entity preservation should be performed in a selective manner,
prioritizing data sources and properties according to their topical relevance for the Web
archive, general importance as well as quality (e.g. in terms of relative completeness,
mutual consistency, and freshness). As such metadata is rarely available, methods to
generate this metadata need to be developed.
An important consideration when crawling LOD sources is the decision when to
stop the crawl. A natural stopping criterion is when all linked resources are crawled,
but because of the scale of the Linked Open Data Cloud this is neither feasible nor
67
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
8
desirable. In Web crawling it is common to restrict the crawl to specific depth, i.e. only
crawl paths of a given maximum length. This model can be generalized as a cost-based
model, where each link traversal is assigned a cost and the crawler only follows paths
where the sum of link costs is less than a specified budget. In this model we can assign
costs to links based on their expected utility for the archive. For example, commonly
used properties such as rdfs:label are typically useful because they are understood by
more applications. By using heuristics we can dynamically prioritize which links to
follow and limit the crawl to the most relevant properties.
Entity-centric crawling: Entities in Linked Data Cloud can be retrieved from SPARQL
endpoints, through URI lookups or from data dumps [14]. The entity preservation pro-
cess can be a part of a Web archive metadata enrichment that extracts entities from the
archived Web pages and fed them into the Linked Data Crawler (e.g. [15]) for preser-
vation. Figure 2(b) present the overall procedure for entity extraction and crawling. In
total, Entity-centric crawling includes following steps:
Step 2a. For each entity extracted by NER, collect entity representations from the rele-
vant datasets. In this step the crawler collects datatype properties (labels) and object
properties (URIs) of a given entity typically using SPARQL queries. To form the
queries, labels, types and (if available) initial URIs resulting from the NER process
as well as available schema information can be used (see e.g. [9]).
Step 2b. Collect entity view(s) available through the HTTP protocol. Here the crawler
can use Web crawling techniques to collect Web pages containing human-readable
entity representations (see Figure 1 (a)).
Step 2c. Follow object properties (URIs) of an entity to collect related entities from the
same LOD dataset. Here, property weighting together with other heuristics (e.g.
path length) can be used to prioritise the crawler.
Step 2d. Follow the links (e.g. owl:sameAs) to external datasets to collect equivalent or
related entities. In this step, prioritization of the crawler can be performed based on
the estimated data quality parameters in such external datasets.
Using the steps above, the crawler can collect comprehensive representations of en-
tities and their sources for archiving. In the next section we describe models to preserve
this data.
5 Data Models for Preservation
In order to model entities within a Web archive taking into account content, schema,
provenance, quality, authenticity and context dimensions, we need to preserve not only
the entities, but also the metadata of the sources these entities originate from. Therefore,
in this section we briefly present the Entity Model and Entity Source Model that
jointly address these requirements.
In order to serialize the models presented below, an RDF representation making use
of existing vocabularies (e.g. RDF Schema5 , VoID6 for data source schemas and PROV-
5
http://www.w3.org/TR/rdf-schema/
6
http://www.w3.org/TR/void
68
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
9
O7 for provenance properties) as well as an icrawl specific vocabulary8 can be created
and written to an RDF file or directly to a triple store within the Web archive. The data
models discussed in this paper describe properties that become necessary based on the
requirements described in Section 3. These models can be extended according to the
needs of particular archival applications.
5.1 Entity Model
According to the requirements identified in Section 3, in the Entity Model, we cover
the aspects of entity content, schema, provenance, authenticity and context. Our entity
model is shown in Table 1.
First of all, the Content and Schema properties give an overview of the entity content
in its original dataset. With respect to the content we differentiate between the key
properties such as label and type and other properties. More detailed property weighting
is provided as a part of Entity Source Model.
Then, Provenance and Quality properties include unique entity identifiers within the
Web archive and its original source (e.g. a URI in the LOD source), as well as content
verification properties. Furthermore, they provide information on entity relevance in its
original context.
Finally, Authenticity and Context properties link the entity to its human-readable
representation within the Web archive and specify important relationships to other enti-
ties within the archive (e.g. relationships between the entities extracted from documents
and entities crawled from LOD, or equivalence relationships with other entities).
5.2 Entity Source Model
The Entity Source Preservation Model that represents a specific entity source is shown
in Table 2. The Content and Schema properties of the Entity Source Model include
domain and schema information as well as detailed property weighting to facilitate
prioritization of the crawler.
The Provenance and Quality properties include identifier of the data sources in-
cluding their URLs, versions and access times, as well as quality parameters and legal
aspects.
6 Related Work
The problems discussed in this paper are particularly relevant to the fields of Web
archiving and Web archive enrichment, Linked Data crawling, metadata dictionaries,
as well as entity dynamics and entity interlinking. In this section we discuss related
work in these areas in more detail.
Web archiving: The goal of Web archiving [18] is to preserve Web pages and sites
7
http://www.w3.org/TR/2013/REC-prov-o-20130430/
8
http://l3s.de/icrawl-ns#
69
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
10
Property Description Properties for
RDF Serialization
Content and Schema
label A human-readable entity name. rdfs:label
type An entity type (according to the original rdf:type
source, or a NER extractor) e.g. Person, Lo-
cation, Organization.
properties Properties as specified in the original taken from the source
source.
Provenance and Quality
URI Persistent identifier of the entity within the
Web archive
version identifier Unique identifier for each archived entity icrawl:hasVersion
version.
entity source A reference to the entity source (see Entity void:Dataset
Source Model).
origin URI The URI of the entity in the original source. prov:hadPrimarySource
crawl timestamp Timestamp when the entity was extracted or prov:generatedAtTime
retrieved.
fingerprint Verification of the content, including algo- icrawl:hasFingerprint,
rithm and value.
prov:wasGeneratedBy, rdf:value
software Software used to extract or retrieve the en- prov:SoftwareAgent
tity.
relevance score Relevance score of the entity including an icrawl:hasAlgorithm,
algorithm
(e.g. ObjectRank[2] or frequency) and the prov:wasGeneratedBy, rdf:value
score.
Authenticity and Context
visualization URI A unique identifier of the visual representa- icrawl:visualizedBy
tion of the entity within the Web archive.
snippet A short textual representation of the entity dc:description
using e.g. surrounding sentences or descrip-
tive properties.
relation Relations to other objects in the Web e.g. owl:sameAs
archive, e.g. equivalence.
Table 1. Entity Model
70
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
11
Property Description Properties for
RDF Serialization
Content and Schema
domain A topical domain of the dataset rdfs:domain
class Relevant entity types in the dataset rdfs:Class
schema
weighted properties Weighted properties for archiving priori- icrawl:hasProperty,
tization.
The weight is in the range [0,1], where rdf:property, rdf:value
“1” stands for the highest archiving pri-
ority.
Provenance and Quality
URI Persistent identifier of the source within
the Web archive.
origin URI URL of the source on the Web at the void:uriLookupEndpoint
archiving time point.
version identifier Unique and stable identifier for each ver- icrawl:hasVersion
sion of the source.
quality parameters Quality metrics (e.g. data size, freshness, e.g. void:entities, void:properties
completeness of data).
legal parameters Publisher and license of the source. dc:publisher, icrawl:license
Table 2. Entity Source Model
71
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
12
for the future. This is required because Web pages disappear rapidly [25] or change their
content. Therefore it is necessary to preserve their content in Web archives, of which
the most prominent is the Internet Archive9 .
Web archives are typically created using Web crawlers (e.g. Heritrix [20]) which fol-
low hyperlinks between Web pages and store the content of Web pages they encounter
on the way. Depending on the policies of the archiving institution, links are followed in
the order they are encountered (breadth-first-crawling) or are prioritized according to
their relevance for a given crawl topic (focused crawling [4]). In this paper we pay spe-
cific attention to the aspects of entity-centric Linked Data crawling for Semantic Web
resources. These aspects has not been addressed sufficiently in the related work.
Web archive enrichment: Linked Open Data has been successfully used for Web
archive enrichment in the ARCOMEM project10 . To this extent, entity representations
extracted from textual documents using NER techniques have been linked with the
corresponding entities in DBpedia and Freebase datasets[9]. This mapping enabled co-
resolution of entity representations extracted from multiple Web documents. However,
the LOD entities used for the co-resolution and Web archive enrichment have not been
preserved. As a consequence, external links to the dynamic Freebase dataset became
obsolescent very quickly. The entity preservation techniques discussed in this paper
can alleviate this problem by providing a stable time-based representation of linked
entities within the Web archive.
Linked Data crawling and versioning: Recently, there have been many studies in
the areas of Linked Data crawling [15], Semantic Web search engines (e.g. Sindice11 )
as well as distributed SPARQL querying, e.g. [14]. The Memento project12 aims at
providing an HTTP-based versioning mechanism for Web pages and Linked Data and
implements versioning of Linked Data as a part of content negotiation [8]. However,
existing solutions do not provide entity-centric preservation facilities. In contrast, we
collect Linked Data resources in a focused way by paying specific attention to their
relevance in the context of Web archives.
Metadata dictionaries: Multiple models and metadata dictionaries support descrip-
tions of entities and their specific properties at different level of details and can po-
tentially be useful in the context of entity preservation. For example, PROV Family of
Documents13 comprises a set of W3C recommendations for representing provenance
information and VoID 14 is an RDF Schema vocabulary for expressing metadata about
RDF datasets, just to name some examples. While some properties from the existing
models and data dictionaries can be mapped to the properties of the data models pro-
posed in this paper, they do not cover the entire spectrum of preservation aspects of
9
https://archive.org/
10
www.arcomem.eu
11
http://sindice.com
12
http://mementoweb.org
13
http://www.w3.org/TR/prov-overview/
14
http://www.w3.org/TR/void/
72
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
13
distributed entities in LOD. To ensure interoperability, we reuse existing properties in
the model serialization, where applicable.
Entity dynamics: The detection of the dynamics of entities attracted recently a
number of researchers with a special focus on query reformulation. Berberich et al. [3]
propose reformulating a query into terms prevalent in the past by measuring the de-
gree of relatedness between two terms when used at different times by comparing the
contexts as captured by co-occurrence statistics. Kanhabua and Nørvåg [17] incorpo-
rate Wikipedia for detecting former names of entities. They exploited the history of
Wikipedia and consider anchor texts at different times pointing to the same entity as
time-based synonyms. Kaluarachchi et al. [16] propose to discover semantically iden-
tical concepts (or named entities) used at different time periods using association rule
mining to associate distinct entities to events. Tahmasebi et al. [26] proposed an un-
supervised method for named entity evolution recognition in a high quality newspaper
(i.e., New York Times). They consider all co-occurring entities in change periods as
potential succeeding names and filter them afterwards using different techniques.
Entity interlinking: Entity interlinking identifies equal, similar or related entities
[10, 23] and formally describes such relations. The state-of-the-art in link discovery
is mostly focused on owl:sameAs relations linking entities representing the same real-
world objects in distributed data collections e.g. [21, 22]. In the context of LOD preser-
vation, we retrieve entities from distributed sources independently. To perform Web
archive consolidation, we can rely on state-of-the-art entity interlinking methods and
available equivalence links.
7 Conclusions and Outlook
The contributions of this paper are manyfold. First of all, we described use cases for en-
tity preservation in Linked Open Data in the context of Web archives. These use cases
include applications for entity-based access to long term archives, evolution aware en-
tity search and time sensitive access to LOD. Then we derived requirements for entity
preservation with respect to the content, schema, provenance, quality, authenticity and
context dimensions. Following that, we discussed methods for entity extraction and
preservation in Web archives and presented data models to represent data sources and
entities within Web archives according to the identified dimensions. As the next steps
towards the entity-centric LOD preservation we envision investigation of preservation-
relevant quality aspects of Linked Open Datasets, as well as development of the meth-
ods to automatic property weighting to achieve effective prioritization in the entity
crawling process.
Acknowledgments
This work was partially funded by the European Research Council under ALEXANDRIA (ERC
339233).
73
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
Bibliography
[1] R. Al-Rfou’ and S. Skiena. Speedread: A fast named entity recognition pipeline. In COL-
ING, pages 51–66, 2012.
[2] A. Balmin, V. Hristidis, and Y. Papakonstantinou. Objectrank: Authority-based keyword
search in databases. In Proceedings of the Thirtieth International Conference on Very Large
Data Bases - Volume 30, VLDB ’04, pages 564–575. VLDB Endowment, 2004. ISBN 0-12-
088469-0. URL http://dl.acm.org/citation.cfm?id=1316689.1316739.
[3] K. Berberich, S. J. Bedathur, M. Sozio, and G. Weikum. Bridging the terminology gap in
web archive search. In WebDB, 2009.
[4] S. Chakrabarti, M. van den Berg, and B. Dom. Focused crawling: a new approach
to topic-specific web resource discovery. Computer Networks, 31(11-16):1623–1640,
1999. ISSN 1389-1286. doi: 10.1016/S1389-1286(99)00052-3. URL http://www.
sciencedirect.com/science/article/pii/S1389128699000523.
[5] X. Cheng and D. Roth. Relational inference for wikification. In EMNLP, 2013. URL
http://cogcomp.cs.illinois.edu/papers/ChengRo13.pdf.
[6] R. Collobert, J. Weston, L. Bottou, M. Karlen, K. Kavukcuoglu, and P. Kuksa. Nat-
ural language processing (almost) from scratch. J. Mach. Learn. Res., 12:2493–2537,
Nov. 2011. ISSN 1532-4435. URL http://dl.acm.org/citation.cfm?id=
1953048.2078186.
[7] M. Cornolti, P. Ferragina, and M. Ciaramita. A framework for benchmarking entity-
annotation systems. In Proceedings of the 22Nd International Conference on World Wide
Web, WWW ’13, pages 249–260, Republic and Canton of Geneva, Switzerland, 2013. In-
ternational World Wide Web Conferences Steering Committee. ISBN 978-1-4503-2035-1.
URL http://dl.acm.org/citation.cfm?id=2488388.2488411.
[8] H. V. de Sompel, R. Sanderson, M. L. Nelson, L. Balakireva, H. Shankar, and S. Ainsworth.
An http-based versioning mechanism for linked data. CoRR, abs/1003.3661, 2010. URL
http://arxiv.org/abs/1003.3661.
[9] S. Dietze, D. Maynard, E. Demidova, T. Risse, W. Peters, K. Doka, and Y. Stavrakas. Preser-
vation of social web content based on entity extraction and consolidation. In 2nd Interna-
tional Workshop on Semantic Digital Archives (SDA) in conjunction with the 16th Inter-
national Conference on Theory and Practice of Digital Libraries (TPDL), Pafos, Cyprus,
September 2012, 2012.
[10] A. K. Elmagarmid, P. G. Ipeirotis, and V. S. Verykios. Duplicate record detection: A survey.
IEEE Trans. on Knowl. and Data Eng., 19(1):1–16, Jan. 2007. ISSN 1041-4347. doi:
10.1109/TKDE.2007.9. URL http://dx.doi.org/10.1109/TKDE.2007.9.
[11] P. Ferragina and U. Scaiella. Tagme: On-the-fly annotation of short text fragments (by
wikipedia entities). In Proceedings of the 19th ACM International Conference on In-
formation and Knowledge Management, CIKM ’10, pages 1625–1628, New York, NY,
USA, 2010. ACM. ISBN 978-1-4503-0099-5. doi: 10.1145/1871437.1871689. URL
http://doi.acm.org/10.1145/1871437.1871689.
[12] J. R. Finkel, T. Grenager, and C. Manning. Incorporating non-local information into infor-
mation extraction systems by gibbs sampling. In Proceedings of the 43rd Annual Meeting
on Association for Computational Linguistics, ACL ’05, pages 363–370, Stroudsburg, PA,
USA, 2005. Association for Computational Linguistics. doi: 10.3115/1219840.1219885.
[13] G. Gossen, E. Demidova, T. Risse, and G. B. Tran. Towards entity-centric preservation
for web archive enrichment. In Joint Proceedings of the USEWOD ’15 and the PRO-
FILES’15 Workshops, Portorož, Slovenia, May 31 - June 1, 2015., volume 1362 of CEUR
74
Proceedings of the 5th International Workshop on Semantic Digital Archives (SDA 2015)
15
Workshop Proceedings, pages 81–84. CEUR-WS.org, 2015. URL http://ceur-ws.
org/Vol-1362/PROFILES2015_paper5.pdf.
[14] A. Harth, K. Hose, M. Karnstedt, A. Polleres, K.-U. Sattler, and J. Umbrich. Data sum-
maries for on-demand queries over linked data. In Proceedings of the 19th International
Conference on World Wide Web, WWW ’10, pages 411–420, New York, NY, USA, 2010.
ACM. ISBN 978-1-60558-799-8. doi: 10.1145/1772690.1772733.
[15] R. Isele, J. Umbrich, C. Bizer, and A. Harth. LDSpider: An open-source crawling frame-
work for the web of linked data. In Proceedings of 9th International Semantic Web Confer-
ence (ISWC 2010) Posters and Demos, 2010.
[16] A. C. Kaluarachchi, A. S. Varde, S. J. Bedathur, G. Weikum, J. Peng, and A. Feldman.
Incorporating terminology evolution for query translation in text retrieval with association
rules. In CIKM, pages 1789–1792, 2010.
[17] N. Kanhabua and K. Nørvåg. Exploiting time-based synonyms in searching document
archives. In Proceedings of the 10th annual joint conference on Digital libraries, JCDL
’10, pages 79–88, New York, NY, USA, 2010. ACM.
[18] J. Masanés. Web Archiving: Issues and Methods, pages 1–53. Springer, 2006. ISBN 978-
3-540-23338-1. doi: 10.1007/978-3-540-46332-0 1.
[19] D. Milne and I. H. Witten. An open-source toolkit for mining wikipedia. Artif. Intell., 194:
222–239, Jan. 2013. ISSN 0004-3702. doi: 10.1016/j.artint.2012.06.007.
[20] G. Mohr, M. Kimpton, M. Stack, and I. Ranitovic. Introduction to Heritrix, an archival
quality web crawler. In 4th International Web Archiving Workshop (IWAW04), 2004.
[21] A. Nikolov, M. d’Aquin, and E. Motta. Unsupervised learning of link discovery config-
uration. In The Semantic Web: Research and Applications - 9th Extended Semantic Web
Conference, ESWC 2012, Heraklion, Crete, Greece, May 27-31, 2012. Proceedings, vol-
ume 7295 of Lecture Notes in Computer Science, pages 119–133. Springer, 2012. ISBN
978-3-642-30283-1.
[22] G. Papadakis, E. Ioannou, C. Niederée, T. Palpanas, and W. Nejdl. Eliminating the redun-
dancy in blocking-based entity resolution methods. In Proceedings of the 2011 Joint Inter-
national Conference on Digital Libraries, JCDL 2011, Ottawa, ON, Canada, June 13-17,
2011, pages 85–94, 2011.
[23] G. Papadakis, E. Ioannou, C. Niederée, T. Palpanas, and W. Nejdl. Beyond 100 million
entities: large-scale blocking-based resolution for heterogeneous data. In Proceedings of
the fifth ACM international conference on Web search and data mining, WSDM ’12, pages
53–62, New York, NY, USA, 2012. ACM. ISBN 978-1-4503-0747-5.
[24] L. Ratinov and D. Roth. Design challenges and misconceptions in named entity recogni-
tion. In Proceedings of the Thirteenth Conference on Computational Natural Language
Learning, CoNLL ’09, pages 147–155, Stroudsburg, PA, USA, 2009. Association for Com-
putational Linguistics. ISBN 978-1-932432-29-9.
[25] H. M. SalahEldeen and M. L. Nelson. Losing my revolution: How many resources
shared on social media have been lost? In Theory and Practice of Digital Libraries,
volume 7489, pages 125–137. Springer, 2012. ISBN 978-3-642-33289-0. doi: 10.1007/
978-3-642-33290-6 14.
[26] N. Tahmasebi, G. Gossen, N. Kanhabua, H. Holzmann, and T. Risse. Neer: An unsuper-
vised method for named entity evolution recognition. In Proceedings of the 24th Interna-
tional Conference on Computational Linguistics (Coling 2012), Mumbai, India, 2012. URL
http://www.l3s.de/neer-dataset/.
[27] R. Verborgh, M. V. Sande, P. Colpaert, S. Coppens, E. Mannens, and R. V. de Walle.
Web-scale querying through linked data fragments. In 7th Workshop on Linked
Data on the Web. CEUR series of workshop proceedings, 2014. URL http://
linkeddatafragments.org/publications/.
75