This paper addresses the issue of representing time entities (i.e. instants and intervals) as Linked Data, and how to exploit topological temporal relationships in order to increase the connectivity degree within Linked Data sets. Describing and e ciently managing temporal information in knowledge management systems is important. Information is volatile, dependant on a number of contexts for its interpretation, among them \time". Many data sets contain information that is valid only within a given time frame (e.g. roles ful lled by di erent people at di erent times), whereas others describe temporal events. In this paper we present an approach to describe temporal entities as reusable URIs that can be adopted by data publishers as a temporal context for their information resources. The approach identi es a set of discrete temporal entities as relevant for a certain domain (e.g. nancial years for the public sector) while a RESTful API is provided to users to dynamically create their own temporal entities. Once a dynamic temporal URI is resolved, information is provided to situate such URI in reference to the domain relevant entities. The URI resolution employs simple topological temporal reasoning in order to exploit the qualitative relationships between entities. We also provide a usage scenario of our approach based on a backlinking service and using Public Sector Information published in Linked Data format within the EnAKTing project.
The Linked Data initiative represents the rst collaborative e ort to create a
Web of Data (WoD henceforth) at scale, providing a few, simple guidelines for
publishing content using well established standards [
The nature and validity of the information is often related to a time frame and is therefore not universal. For example, the de nition of a constituency is temporary (e.g. Southampton Test constituency was established in 1950, the date where the previous Southampton constituency ceased to exist). The representation of temporal entities within the Linked Data cloud is therefore an essential step in order to provide temporal context to other entities.
A common trait of PSI, composed largely of national statistics, is its temporal validity (i.e. the time frame when the data was collected). Having a continuous collection of data helps policy analysts study trends, but it requires a coherent representation of time entities in the information ecosystem.
In this paper, we present a pattern and a tool for representing time lines that
support the temporal contextualization of linked data entities. Such a pattern
allows us to describe qualitative relations among temporal entities as well as
absolute time points and intervals that can then be queried and exploited in
order to retrieve relevant resources. Allen's temporal intervals algebra [
Time representation, querying and reasoning is a well known topic in computer
science where many proposals have been produced; from formal systems, to
ontologies and query languages. Within the database community the representation
and management of temporal information can be found for example in the
temporal extension to SQL named TSQL2 [
Allen provided a model of time based on intervals [
Such a formalization of time has been encoded, as reported above, in an OWL
ontology [
X before Y X
Y
X equal Y
X Y
X meets Y X
Y
X overlaps Y X
Y
X during Y
X Y
X starts Y X Y
X finishes Y
X Y between Time Entity instances. Time Instants, although not present in the original Allen model, are treated as point form intervals whose starting and ending point coincide.
Temporal validity of RDF statements was originally investigated by
Gutierrez et al. [
Contextualization in the WoD relies on authoritative sources of URIs for
naming entities. A typical example of this usage pattern can be seen by looking
at DBpedia [
Creation of URIs for describing reference time intervals has been proposed
rstly by Ian Davis with his http://placetime.com site where he uses ISO 8601
standard [
In this paper we present a pattern for describing temporal information within Linked Data. This approach allows users to refer to reference time intervals sup
accessed at 17/05/2010 plied by data providers (e.g. quarter of nancial years) as well as unanticipated temporal entities within a time line. The resolution of such dynamic URIs gives not only useful information about the time entity itself (like that provided by http://placetime.com) but also temporal topological relationships in reference to the managed discrete time entities.
Temporal reasoning adopted here is a proper subset of Allen's classical time algebra, namely only topological temporal relationships are taken into account in order to retrieve relevant URIs. The use of backlinks, jointly with a lightweight temporal query support, allows us then to retrieve the contextualised resource. 3
The Linked Data principles [
The RDF data model inherits XML schema (xsd henceforth) support for data types (i.e. xsd:date and xsd:dateTime). However, the sole use of xsd without the adoption of a linkable URI representation for relevant time entities, diminishes the level of connectivity of the overall data cloud, making the information contextualization process more di cult. xsd time related data types provide in fact a uniform representation of the data semantics whose interpretation is attened into a shared time line, regardless of the context of use of the time information. Moreover, thinking of the WoD as a hypermedia system browsable by means of software agents, the node connectivity is as important as the schema that provides nodes with semantics.
For instance, let us consider the concept of a nancial year. Although it is
present in many countries, its actual extension di ers from country to country,
but even if the extension of the nancial year in the UK and in India is the same
(both of them start in April), it would be quite unsettling asserting that they are
the same thing. Considering the semantics of nancial year merely as a product
of its extensional features (i.e. starting and ending date), this sentence would
lead to no semantic clash at all. However, from an information retrieval point
of view, binding UK relevant data to the Indian nancial year, still seems quite
inaccurate and misleading. In order to refer to di erent temporal entities we
need to lift up from a at representation of time and create distinct name spaces
that restore contextual di erences and identities. In the previous example, this
would lead to the creation of di erent URIs for the UK and Indian nancial
year and then to the correct link between the UK relevant information and the
UK temporal context. The concomitant creation of contextually equivalent name
spaces for temporal entities only defers the creation of authoritative sources of
URIs. Co-reference systems [
Another issue that presents itself when dealing with temporal dimensions in publishing and dealing in general with statistical data is the heterogeneous temporal de nition used when collecting observations. Data observations are in fact provided (explicitly or not) for di erent timespans and with di erent granularity (by solar or nancial year, quarter etc.). It is therefore necessary to provide reasoning services for reconciling when possible these di erences facilitating a semantically enabled retrieval of information.
As an example, consider two data sets from http://data.gov.uk about NHS performance3 and mortality4. These data sets have been translated into Linked Data entities but unfortunately, the rst data was described by quarter of nancial year while the latter by whole nancial years. Either at retrieval and at aggregation phase, the containment information between temporal entities must be explicitly represented and exploited in order to produce meaningful results. As an example, the knowledge that (i) nancial years are composed by four quarters, that (ii) each of them is temporally contained by it, and that (iii) statistical observation can be summed up when aggregating to supersets of dimensions can be exploited for producing aggregate statistics that were not originally given.
Furthermore, considering more broadly general Linked Data sets, if we are to represent time validity of entities linking them with temporal entities, we would be able to exploit temporal entity semantics.
For an e ective description of time that could help the above mentioned issues, the following high level requirements must be satis ed: explicit URI representation of temporal entities of general interest, and support for some form of temporal reasoning.
In the rest of the paper we will describe a proposal for the description and management of temporal URIs for Linked Data entities. 4
In order to allow users to explicitly express temporal facets of their data reusing reference URIs we developed the concept of Linked Timelines, knowledge bases about general instants and intervals that expose resolvable URIs. Linked Timelines adopt the OWL time ontology as the standard vocabulary for describing temporal entities. Temporal entities in a timeline contain RDF statements not only de ning their starting and nal instant, but also temporal relationships between other entities managed in the timeline creating a lattice of temporal entities. Such information will be used in order to infer temporal topological information as will be described in this section.
The OWL Time ontology5 has been extended with three datatype properties: hasXSDStart, hasXSDEnd, and hasXSDDuration for describing with XML Schema datatype xsd:dateTime and xsd:duration the starting and ending instant of a time interval and its duration respectively. These extensions act as a short-cut for de ning temporal intervals without creating rst the instances for the starting and ending instants. Intervals and not Instants themselves are in fact
the main objects represented within timelines. Therefore the start-end instants would either be anonymous nodes, which are to be avoided whenever possible when publishing linked data, or verbose second class instances that would clutter the knowledge base. Such short-cut has been provided in the OWL Time ontology for the Time Instants with the creation of the inXSDDateTime datatype property, but not for time intervals. We introduced the possibility of de ning intervals reusing entirely XML schema datatypes in order to limit the creation of entities to the ones of actual interest. An example of a temporal entity described within a Linked Timeline can be seen in Figure 2 where the RDF description for the solar year 2007 is reported in Turtle syntax. Note that in Figure 2 there is no explicit usage of the property hasXSDEnd, but its value can be inferred by exploiting the semantics of time:intervalMeets property and knowing the starting instant of <http://time.psi.enakting.org/id/2008> (i.e. the solar year 2008). <http://time.psi.enakting.org/id/2007> a timepsi:Year ; rdfs:label "Solar year 2007" ; timepsi:hasXSDStart "2007-01-01T00:00:00Z"^^xsd:dateTime ; timepsi:hasXSDDuration "P1Y"^^xsd:duration ; time:intervalContains <http://time.psi.enakting.org/id/2007/08/Q1> , <http://time.psi.enakting.org/id/2007/08/Q2> , <http://time.psi.enakting.org/id/2007/08/Q3> ; time:intervalStartedBy <http://time.psi.enakting.org/id/2006/07/Q4> ; time:intervalMeets <http://time.psi.enakting.org/id/2008> ; time:intervalMetBy <http://time.psi.enakting.org/id/2006> .
The http://time.psi.enakting.org timeline contains relevant URIs for the UK nancial year (other domains could describe di erent kinds of temporal entities), therefore alongside solar years we nd nancial years and quarters of nancial year. Other timelines can be set up in order to represent time entities relevant to other domains and applications. The RDF representation of the solar year 2007 in Figure 2 for example, additionally to the starting date and duration, contains the relations between the year 2007 and other relevant intervals such as the the solar year 2006 (that meets 2007) and 2008 (met by the 2007). The relevance criteria is determined by the data publisher and, in this example, the relevant entities are the quarters of nancial years contained in the year 2007 (i.e. the fourth quarter of the nancial year 2006/07 and the rst three quarters of the nancial year 2007/08).
In order to limit the size of the knowledge base we do not describe before and after relationships between every instance, which will cause a geometric explosion in the number of statements. Instead we limited the kind of relationships explicitly described to the topological ones (i.e. time:intervalContains and time:intervalDuring) and the ones useful to recreate the lattice of entities (i.e. time:intervalMeets, time:intervalMetBy, time:intervalStarts, and time:intervalStartedBy). It is noteworthy that in Figure 2 we did not represent the overlapping relationships between the solar year 2007 and the nancial year 2007. In the UK in fact, the nancial year starts the rst of April of the solar year, therefore every solar year overlaps its nancial year.
The rational for limiting the kind of relationships to the topological ones is
twofold: primarily if all the temporal relationships were to be reported, as stated
before, there could be a potential problem of space complexity. Secondarily, a
topological relationships' semantics is easier to model when dealing with data
integration and aggregation. In fact, as argued by [
The space of discrete URIs described so far is complemented by a RESTful interface for de ning dynamically time instants and intervals that happen in the same timeline. When resolving such dynamic URIs a document is returned that provides a description of the instant/interval. The format of the document is either decided via a content negotiation mechanism or by directly stating the desired format. Encoding time entities via URIs allows users to refer explicitly to particular timelines (e.g. for UK nancial periods) even if the actual URIs are not known or not yet existent, enabling therefore a future evolution of the timeline.
The API interface for creating dynamic temporal URIs follows the http: //placetime.com encoding6, and it is de ned as follows: http://time.psi.enakting.org/{type}/{YYYY}-{MM}-{DD}T{hh}:{mm}:{ss}{TZ} P[{y}Y][{m}M][{d}D][T[{h}H][{n}M][{s}S]][/{format}] Where the rst mandatory part is de ned as follows: { type is either interval or instant { YYYY Four digits (from 0001 to 9999) represent the year (only AD dates are valid). { MM Two digits (from 01 to 12) represent the month of the year. { DD Two digits (from 01 to 28, 29, 30 or 31 depending on the month) represent the day of the month. { hh Two digits (from 00 to 23) representing the hour of the day. { mm Two digits (from 00 to 59) representing the minute of the hour. { ss Two digits (from 00 to 59) representing the second of the minute. { TZ A string representing the timezone. This can be either a Z for UTC timezone or a string in the format: [+/-]hh:mm representing an o set of hh hours and mm minutes from the UTC timezone.
If type is interval, then a duration must be given and least one of the following parameters must be provided: { y A number of years in the interval { m A number of months in the interval { d A number of days in the interval { h A number of hours in the interval { n A number of minutes in the interval { s A number of seconds in the interval
The last part is optional and de nes the format of the returned document: { format, if present, can be one of: doc, rdf, or ttl and de nes the format of the document returned (HTML, RDF/XML, or Turtle respectively). The content negotiation will return the same document if this parameter would be missing and the Accept: header in the HTTP request would be set to the wanted MIME type (e.g. application/x-turtle for the ttl format).
As an example of usage of this API, if we consider the following instant URI: http://time.psi.enakting.org/instant/2007-01-10T10:00:00Z/ttl, it will return the following Turtle document: @PREFIX timepsi: <http://time.psi.enakting.org/def/> . <http://time.psi.enakting.org/instant/2007-01-10T10:00:00Z> a time:Instant ; rdfs:comment "The instant 10:00:00Z, of the tenth day of the month of January in the year 2007 of the
Gregorian calendar." ; timepsi:hasXSDStart "2007-01-10 10:00:00+00:00"^^xsd:dateTime ; rdfs:label "2007-01-10T10:00:00+00:00" ; time:inside <http://time.psi.enakting.org/id/2007> , <http://time.psi.enakting.org/id/2006/07> , <http://time.psi.enakting.org/id/2006/07/Q4>.
As per the previous example, for de ning intervals, if we consider the following URI: http://time.psi.enakting.org/interval/2007-01-01T12:00: 00ZP4M/ttl, it will return the Turtle document reported as follow: @PREFIX timepsi: <http://time.psi.enakting.org/def/> . <http://time.psi.enakting.org/interval/2007-01-01T00:00:00ZP4M> a time:ProperInterval ; rdfs:comment "A time-interval of exactly 4 month(s), beginning at 0:00:00Z, on the first day of the month of January in year 2007 of the Gregorian calendar." ; rdfs:label "2007-01-01T12:00:00+00:00P4M" ; timepsi:hasXSDStart "2007-01-01 00:00:00+00:00"^^xsd:dateTime ; timepsi:hasXSDDuration "P4M"^^xsd:duration ; time:intervalDuring <http://time.psi.enakting.org/id/2007> ; time:intarvalContains <http://time.psi.enakting.org/id/2006/07/Q4> .
As illustrated in the two code examples above, the RDF documents describe the temporal entity resolved with a label and a description in natural language. Moreover, the documents describe the topological temporal relationships that hold between the resolved entity and the entities managed in the timeline. The rationale for returning only the topological relationships has been discussed in a earlier part of this section.
By default then, when resolving a dynamic URI in a Linked Timeline, the following entities belonging to it are returned, depending on the kind of temporal entity (i.e. instant or interval): instant: all the temporal intervals that contain the instant encoded by the resolved URI (time:inside property is here used to state that an instant is contained within a temporal interval) interval: all the temporal intervals that contain or are contained by the interval encoded by the resolved URI (time:intervalContains and time:interval During are used here respectively to state that an interval contains or is contained by another interval) d e t a e rr c e s U d e g a n a m g n i T K A n E http://time.psi.enakting.org/interval/2007-01-01T12:00:00ZP4M time:intervalContains
time:intervalDuring http://time.psi.enakting.org/id/2006/07/Q4 time:IntervalDuring
http://time.psi.enakting.org/id/2007
The dynamic resolution of temporal entities, together with the temporal reasoning that puts it in context with the managed ones, allows the reuse of time URIs in foreign data sets. Users can in fact create their URIs programmatically (User created in Figure 3) making reference to a Linked Timeline that will automatically contextualize them with reference to a managed set of temporal entities (EnAKTing managed in Figure 3). 4.2
Within the EnAKTing project, we translated public sector information in Linked Data format, aligning the di erent dimensions whenever possible. Geographical dimensions were aligned to the Ordnance Survey administrative ontology but for temporal dimensions we had to create an ad hoc solution. The Linked Timeline instance created for PSI data is accessible at http://time.psi.enakting.org and it is composed by two sections. The rst section (upper part of Figure 4) is a timeline widget7 that illustrates the discrete entities managed by the timeline. 7 For this we have used the Timeline widget from SIMILE project, http://www.
simile-widgets.org/timeline/. The second part of the interface (bottom part of Figure 4) is composed by two calendars that allow to select the starting and ending day of a temporal interval. Alternatively, the user can check the Time Instant check box and the second calendar will be disabled. While the user chooses the temporal extent of its temporal entity the URI just above the two calendars will change accordingly, creating a URI that can be resolved immediately or copied and paste for later use. 4.3
The PSI Backlinking service provides an access point to retrieve backlinks from
Foreign URIs [
In order to improve the connectivity of our data sets we have aligned the temporal entities in a number of our data sets (the process is still ongoing) to our linked Timeline. Then we have collected the back links from the temporal entities and extended our backlinking service in order to exploit the temporal reasoning. The data sets aligned for this evaluation are three: http: //parliament.psi.enakting.org) is an historical record of the UK parliamentary voting, MPs and their constituencies; http://nhs.psi.enakting.org is a temporal series of NHS performance statistics collected over the years (from 2005 to 2008); and http://education.data.gov.uk is a directory of present and past educational institutions reporting ancillary information (e.g. address, number of pupils, opening and closing date).
The Backlinking and the Linked Timeline services run as separate services and the rst service performs HTTP requests to get the temporal entities from the latter that employ temporal containments (see Figure 3) in order to retrieve all the relevant URIs. Entities returned by the temporal reasoner to the Backlinking service are created by a previous scraping of the data sets and are not usually returned to normal users that will be otherwise cluttered with non contextualized temporal entities. When the backlinking service recognizes a URI from the Linked Timeline namespace it gets the list of contained entities for the input URI and returns the backlinks connected to any URI contained in the temporal interval. For example, consider the URI used in Section 4 for de ning temporal intervals: http://time.psi.enakting.org/interval/2007-01-01T00:00:00ZP4M. Calling the backlink service on this URI will return (as shown in Figure 5) the following backlinks: 3 ministerial o ces, 299194 NHS performance statistic items, and 246 divisions. In this paper we have presented a pattern for temporal URIs exposure and a tool based on Allen's temporal algebra for creating such URIs within Linked Timelines that supports the contextualization task. The usage of HTTP resolvable URIs and the decoupling of the information publishing and retrieval phase make this approach suitable for large scale data publishing infrastructures. Use of external links for contextualization of local information instead of using local statements means that local data querying and retrieval can be eventually delegated to external services.
There are many bene ts to be realised by including or retrospectively adding Linked Timelines to data sets, not least of which are enhanced discovery and querying potential providing linkage and exposure for information that may not previously have been accessible. Moreover, the adoption of Linked Timelines does not preclude data publishers from providing local solutions for temporal indexing and information retrieval.
This work was supported by the EnAKTing project funded by the Engineering and Physical Sciences Research Council under contract EP/G008493/1.