We here present a vision of what needs to be addressed when designing and publishing linked data on the Web. Our approach aims at reducing the amount of incorrect, irrelevant, or redundant content { which can also be seen as pollution in the Web of Data { when publishing linked data. At the foundation lie the design principles adapted from green engineering.We envision a holistic framework that evaluates, along these principles and their respective assessment metrics, datasets from publishers and allows con guration of new validation tools.
The rapid growth of the Web of Data has contributed to the creation of large amounts of linked data that often results in low quality content. For this reason, it is important to investigate the problem of pollution from which the linked data environment may su er. Pollution refers in our case to incorrect, irrelevant, or redundant content, which aggregates low value to users and services that consume these data. Examples include broken links, ambiguous use of owl:sameAs, redundant de nition of vocabularies, multiple URIs for the same resource in a dataset, complex vocabularies that cannot be e ciently reused, uncomprehensible data, unaccessible data, non-maintained data, etc.
We approach the eld of green engineering, since it has long been involved
with quality assurance when designing materials, processes and systems that are
benign to the environment. Moreover, this eld o ers an ecological perspective
when discussing the problems encountered in linked data publishing. At the
foundation of our approach lie the fundamental principles of green engineering [
The issue of quality on the Web of Data has been addressed along aspects
such as syntax errors and inconsistencies in datasets [
We introduce each principle with a short description and assessment measures,
which are partly contribution of Web community [
Ensure that data are as inherently benign as possible Benign refers to data that maximize the qualities in which the publishers and consumers are interested. Publishers are interested that their data is consumed, i.e. data is 1)accessible 2)understandable by consumers and 3)meet their demand. Dimension Measures
It is better to prevent waste than to treat or clean up after it is formed. Publishers should strive to produce data with "zero-waste", which in the Web of Data results from the lack of use or consumption, i.e. consumers (human and machines) are unable to e ectively exploit published data for bene cial use.
Dimension Measures
Modularization operations should be a component of the design process Engineering large monolithic ontologies leads to artifacts that can rarely be reused, due to tting to the design requirements. Modularization helps solve this challenge using instead a set of micro-ontologies, therefore increasing opportunities for the reuse of the developed artifacts.
Dimension Measures
Design datasets in order to maximize e cient exploitation Published linked data should allow consumers to search, query and browse them achieving required results with minimum e ort and time.
Dimension Measures
Bringing content and publishing rate in line with demand Publishers should have possible consumers in mind when designing their data.To this aim, they should cover user needs providing only the necessary resources. Dimension Measures
When making design choices, publishers should strive to reuse a complex ontology or dataset as it is, instead of recycling i.e. extracting parts of it and modifying them for further use. Complexity should be viewed as an investment for reuse. Dimension
Measures
Design for unnecessary capability or capacity solutions should be considered a design aw Publishers should try to provide datasets that meet the necessary capabilities, with no excessive details, while "`one size ts all"' solutions are a design aw.
Dimension Measures
Design for performance in a commercial afterlife It is necessary to provide updates and maintainance after the planned end of life of the data. To reduce waste, components that remain functional and valuable can be recovered for reuse and/or recon guration.
Dimension Measures
In this section, we introduce an envisioned framework, which is a Web platform addressing three main groups of visitors 1) those who want to learn about linked data and the green approach, 2) publishers that wish to check their linked data before publishing them online, and 3) software developers who can contribute with validators that check particular measures pertaining to the principles.
We have initiated the implementation of this framework online1, aiming to make it a future point of reference for the users of the Web of Data. Through the 1 http://www.greenlinkeddata.org introduction of the green principles and the dimensions in which they expand, as well as via the further enrichment of the website with materials and related links, we aim to raise the concern among these users about the importance of the quality of linked data published online.
Besides its informative nature, the platform aims at enabling users to make concious decisions about the data they need to publish, and most importantly help them evaluate how these data conform to the green principles. Therefore, for the publishers the framework o ers the possibility to automatically check the datasets or vocabularies they want to publish based on the measures de ned. A publisher may choose to check its data towards one or several principles and dimensions (Fig. 1).
The evaluation of the data will be done through validators which will consist
of open source or o -the-shelf algorithms o ered in the Web of Data community,
as well as new validators (e.g. to check comprehensibility) that we are
implementing. A more interesting feature of the framework is the possibility provided
to software developers to submit their validators, for example as Web services.
For example one measure for the comprehensability of a dataset is the
labeling completeness metric LClp where lp is a set of labeling properties such as
rdsf:label. This metric evaluates the ratio of non-information resources for
which at least one label is de ned [
There is also the possibility to suggest new dimensions and respectively contribute with appropriate validators. Thus, our goal is to provide an open framework, where Web users not only contribute with validators, but also with new ideas, materials and tools. Furthermore, the platform allows adding to each principle in the website new comments that may consist of, but are not restricted to, best practices, bene ts or even di culties they have had when dealing with those aspects. They may also contribute with suggestions on how to extend dimensions and measures of that principle. 4
At the foundation of this approach lie green engineering principles, which we have transfered to linked data publishing. In contrast to the physical artifacts addressed in the original approach, we deal with data that represent immaterial artifacts. The fundamental di erences between these two types of artifacts have necessarily been taken into account.
Physical artifacts are subject to decay and abrasion in consequence of usage. They cannot easily be duplicated or distributed, and possess the property of excludability. Since material goods are naturally scarce, this can lead to rivalry. In contrast, data are immaterial, thus can be easily duplicated and distributed, without being subject to decay. While porting the principles to the linked data setting, we have extraced only 9 of the original 12 principles, excluding e.g. those dealing with renewable type of resources, infrastructure used to create and provide the data, or discussion on green energy consumption.
In our future work, we will focus on extending the principles with other measures and bringing to life via further development the envisioned framework.