The open world of the (Semantic) Web is a global information space o ering diverse materials of disparate qualities, and the opportunity to re-use, aggregate, and integrate these materials in novel ways. The advent of Linked Data brings the potential to expose data on the Web, creating new challenges for data consumers who want to integrate these data. One challenge is the ability, for users, to elicit the reliability and/or the accuracy of the data they come across. In this paper, we describe a light-weight provenance extension for the voiD vocabulary that allows data publishers to add provenance metadata to their datasets. These provenance metadata can be queried by consumers and used as contextual information for integration and inter-operation of information resources on the Semantic Web.
Whom do you trust? In the human realm, you must earn the trust of others, not assume it. You cannot, for example, expect them to simply hand you their money based on your assurances that you are an honourable person (although they may do that through referral). Conversely, you will give someone else your money or your information only after you have established that they will handle it appropriately (responsibly). And, how do you generate that trust? You generate that trust not by declaring \I'm trustworthy", but by revealing as much information of you as possible.
The above are examples of integration and inter-operation of transactions enabled by notions of quality and trust. Similar examples can be found on the Web. A user, on the Web, may be confronted with a potentially large number of diverse data sources of variable maturity or quality, and selecting the high quality and trustworthy data that are pertinent for their uses and integrating these together may be di cult. The advent of Linked Data brings the potential to expose data on the Web, pointing towards a clear trend where users will be able to easily aggregate, consume, and republish data. It is therefore necessary for end-users to be able to decide the quality and trustworthiness of information at hand.
While in some cases the answer speaks for itself, in others the user will not be con dent of the answer unless they know how and why the answer has been produced and where the data has come from. Users want to know about the reliability or the accuracy of the data they see. Thus, a data integration system, to gain the trust of a user must be able, if required, to provide an explanation or justi cation for an answer. Since the answer is usually the result of a reasoning process, the justi cation can be given as well as a derivation of the conclusion with the sources of information for the various steps.
Provenance, also known as lineage, describes how an object came to be in its present state, and thus, it describes the evolution of the object over time. Provision of the provenance of information resources on the Web can be used as a basis for the assessment of information quality, improving the contextual information behind the generation, transformation, and integration of information on the Web. 2
There are two major research strands of provenance in the literature: data and work ow provenance. In the scienti c enterprise, a work ow is typically used to perform complex data processing tasks. A work ow can be thought of as a set of procedure steps, computer and human, that one enacts to get from the starting state to the goal state. Work ow provenance refers to the record of the entire history of the derivation of the nal output of the work ow. The details of the recording vary from one experiment to another. It may depend on the goals of the experiment, or the regulatory and compliance procedures, and a number of other things. It may involve the recording of the software programs, the hardware, and the instruments used in the experiment.
Data provenance, on the other hand, is more concerned about the derivation of a piece of data that is in the result of a transformation step. It refers to a description of the origins of a piece of data and the process by which it arrives in a database. 3
There are many surveys of existing work on provenance from work ows [
Berners-Lee's \Oh yeah?" button [
The Provenance Vocabulary[
In the linked data world, data are usually collected together and provided as datasets. The provision of the provenance information of datasets' elements is an interesting problem. 4
There are two major challenges in the provision of provenance information of linked data, viz Provenance Representation and Provenance Storage. 4.1
Provenance information can sometimes be larger than the data it describes if the data items under provenance control is ne-grained and the information provided very rich. However, one can reduce storage needs by recording data collection that are important for the operational aspects of the dataset publisher's business.
Provenance can be tightly coupled to the data it describes and located in
the same data storage system or even be embedded within the data le, as
advocated in tSPARQL [
There are two major approaches to representing provenance information, and
these alternate representations have implications on their cost of recording and
the richness of their usages. These two approaches are:
{ The Inversion method: This uses the relationships between the input data,
working backwards, to derive the output data, giving the records of this trace.
Examples include queries and user-de ned functions in databases that can be
inverted automatically or by explicit functions [
While the inversion method is more compact than the annotation approach, the information it provides is sparse and limited to the derivation history of the data. The annotation method, however, provides more information that includes more than the derivation history of the data and may include the parameters passed to the derivation processes, the post-conditions, etc.
We advocate the use of the annotation method as it gives richer information
of the data and the data set we may be interested in. The voiD (Vocabulary of
Interlinked Datasets) [
voidp: Provenance Extension to voiD
In [
nd in existing vocabularies. Reusing existing vocabularies takes advantage of
the ease of bringing together diverse domains within RDF, and it makes data
more reusable. By reusing vocabularies, the data is no longer isolated nor locked
within a single context designed for a single use. We adhered to this advice and
have made use of the following ontologies:
{ Provenance Vocabulary[
In addition, the namespace for voidp is: @prefix voidp: <http://purl.org/void/provenance/ns/>. The classes are: { Actor: Here, we reuse the Actor class in the Provenance vocabulary to specify an entity or an object that performs an action on a particular data item (or a data source or data set), { Provenance: this class is a container class for the list of DataItem(s) we are putting under provenance control, { DataItem: this class models the item of data we put under provenance control.
The properties are: 1. activity: this property speci es that a particular dataset has some items under provenance control, 2. item: speci es the item under provenance control, 3. originatingSource: the item's original source, 4. originatingSourceURI: the URI of the item's original source, 5. originatingSourceLabel: the label text used to describe the item's original source, 6. certification: if the dataset is signed, this property is used to contain the signature elements. This is an important element to prove the origin of a dataset as it is being sliced and diced during its evolution, 7. swp:signature: represents the signature of the dataset, 8. swp:signatureMethod: speci es the signature method, 9. swp:authority: de nes the authority of the relationship between the item under provenance control and the dataset publisher, 10. swp:valid-from and swp:valid-until: these are the valid start and end dates of that (authority) relationship, 11. processType: speci es the type of transformation or conversion procedure carried out on the item's source, e.g. the transformation may be due to some scripts being run on the source data, 12. prv:createdBy: speci es the actor that executes an action on the item that is being recorded. 13. prv:performedAt: date when the transformation is done, 14. prv:performedBy: the URI of the actor that performs the recording of the provenance activity on the item.
These classes and properties are su cient to be useful for specifying information for both work ow and data provenance. 6
Our group, the EnAKTing group2, is dedicated to solving fundamental problems in achieving an e ective web of linked data, and as part of our work, we make use of some of the United Kingdom's government data. As part of our group's work, we recently converted a set of government data les from comma-separatedvalues (csv) to RDF datasets. 6.1
Some of these data les were: { Mortality data:
http://www.statistics.gov.uk/downloads/theme population/ Table 3 Deaths Area Local Authority.xls, { Population data: http://www.statistics.gov.uk/downloads/ theme population/Mid2003 Parl Con quinary est.xls, { Energy:
http://www.decc.gov.uk/assets/decc/statistics/regional/ road transport/ le45728.xls, { CO2 emission: http://www.decc.gov.uk/assets/ decc/ statistics/climate change/ 1 20100122174542 e @@ localregionalco2 emissionsest20057.xls, { Crime: http://www.homeo ce.gov.uk /rds/pdfs09/ hosb1109chap7.xls.
We used voiD to describe these datasets. These datasets and their voiD descriptions were inserted into our RDF database, 4store3. The voiD descriptions used can be found at http://152.78.189.49/voidp/. The provenance elements can be seen in the voiD descriptions.
Example Scenario Query. We may be interested in an example query such as the following:
\Give the originating urls of the datasets for Robbery and female population for the County of Durham in the United Kingdom for 2004. Also give the CO2 emission values and total energy consumption values for that same area. Only give datasets that are from the United Kingdom Home O ce and from the United Kingdom's Department of Energy and Climate Change".
Running such a query, we are given the source urls that were stated in subsection 6.1 (Source Datasets). 7
The provenance of a data element can be used to elicit that datas quality and/or trustworthiness. Data quality can be used as contextual information to aid in data integration. This paper described voidp, a light-weight provenance extension for the voiD vocabulary that allows data publishers to add provenance metadata to the elements of their datasets, enumerating its classes and properties. These provenance metadata can be used by a data integration system or consumer for data aggregation and inter-operation. 8
This work was supported by the EnAKTing project, funded by EPSRC project number EP/G008493/1.