The White House's Materials Genome Initiative (MGI) seeks to substantially improve the process of materials discovery and development, and shorten the time to deployment. One of the main goals of MGI is to develop solutions which provide broader access to scienti c data. This allows materials scientists to integrate each other's data and facilitate communication among scientists working in di erent stages of the materials development continuum.

A key challenge in data integration is dealing with the heterogeneity of data in the Semantic Web community [ 5 ]. Standardized vocabularies are widely used as a component of a toolkit to solve data heterogeneity issues. They play the role of a shared language, which facilitates easy communication and information exchange among people within the community.

While there exist disparate sets of vocabularies developed for the materials domain, there is no easy mechanism to curate these vocabularies by the domain scientists spreading all over the world. The lack of such a mechanism prevents the wider adoption of these vocabularies. Further, existing vocabularies lack the support to capture provenance metadata. Provenance metadata is crucial for data integration from disparate sources in order to determine the trustworthiness of the data and also to give proper credit to the creators of the data. Provenance metadata would increase interoperability, discoverability, reliability as well as reproducibility for scienti c discourse and evidence-based knowledge discovery [ 7 ].

Crowd-sourcing is a cost-e ective and reliable approach to easily distribute a task among a potentially large group of contributors. The Semantic Web community has used crowd-sourcing techniques for knowledge acquisition tasks, including vocabulary development [ 15 ]. Semantic Mediawiki [ 10 ] is one such platform that can be used for crowd-sourced vocabulary curation. However, it lacks built-in support to capture the provenance metadata of RDF triples.

To the best of our knowledge, Wikidata is the only crowdsourced knowledge acquisition platform which supports incorporating such provenance metadata [ 6 ]. It is being used to create and curate structured online database for Wikipedia. Wikidata allows editors to annotate attribute-value pairs using quali ers and references as a way to support metadata. For this purpose, Wikidata uses an auxiliary node in a way which is similar to a blank node as discussed in [ 6 ]. However, the RDF representation of this data model is not intuitive and does not map to the standard RDF triples.

In this work, we try to address the aforementioned challenges with our tool, KnowledgeWiki. We adopt the singleton property template approach developed by Nguyen et al. A singleton property is de ned as a unique property instance representing a newly established relationship between two existing entities in one particular context [ 13 ]. This approach o ers a concise representation for RDF statements about statements, with a formal semantics for an accurate interpretation across applications, tools, and datasets. Recent studies [ 7, 8 ] reported that the singleton property approach o ered the most concise representation on a triple level. Therefore, we adopted the singleton property in our data model for the extension.

In our KnowledgeWiki, by using Semantic MediaWiki as the curation platform and embedding the singleton property approach into Semantic MediaWiki as an extension, we intend to capture metadata of RDF triples such, as provenance information. We combine this extension with other extensions such as semantic forms to provide a user-friendly interface for data collection. The goal of KnowledgeWiki is to provide a wiki-based platform for assisting in the creation and curation of vocabularies in the materials science domain. Our main contributions in this work are three-fold: 1. A Singleton Property Template Extension to Semantic Mediawiki to capture the metadata of RDF triples We incorporate the singleton property approach for RDF data representation into SMW. SMW takes a simple straightforward approach to represent triples. However, as mentioned earlier provenance information is essential, but missing. The singleton property template data model, which is an improvement over the standard rei cation approach, is suitable for representing metadata about the data in materials science [ 7 ]. 2. An algorithm to identify Singleton Property Templates from RDF datasets and to support importing RDF datasets into the wiki We propose an algorithm to automatically identify the regular and singleton property templates for every entity from a given RDF dataset. All the templates related to a single entity will be presented in the same wiki page. This page allows the domain experts to curate the content. 3. Demonstration of the use of this extension in the materials science domain We import three vocabularies extracted from the ASM Handbook Volume 21, MIL-HDBK-5, and MIL-HDBK17 from the materials science domain. We adopted the singleton property template extension for representing the provenance of the vocabularies. The three vocabularies have been curated in our KnowledgeWiki. For representing the provenance information, such as source and license, we reuse the existing vocabularies such as SKOS1, Dublin Core2, and QUDT (Quantities, Units, Dimensions and Data Types)3.

The remainder of the paper is organized as follows: Section 2 discusses the context for the research, Section 3 discusses related work, Section 4 outlines our approach, Section 5 discusses our materials science use case. Finally, Section 6 and Section 7 discusses the future work and conclusion of our work respectively. 2.

The availability of a crowd-sourcing tool is crucial to achieve the requirements proposed by MGI. If made widely available, disparate sources of materials data also could be inventoried to identify gaps in data and to limit redundancy in research e orts. According to MGI, \To bene t from broadly accessible materials data, a culture of data sharing must accompany the construction of a modern materials data infrastructure that includes the software, hardware, and data standards necessary to enable discovery, access, and use of materials science and engineering data. The system should be available to house, search, and curate materials data generated by the community. The initiative also states that the community-developed standards should provide the format, metadata, data types and criteria necessary for interoperability and seamless data integration" [ 19 ].

The Air Force Research Laboratory (AFRL) proposed to work towards the goals of the Materials Genome Initiative, mainly in the context of bringing the materials science community together to collaboratively create and curate a consolidated vocabulary for materials science. For this purpose, AFRL and its partners provided us with initial legacy data. This data is comprised of materials science dictionary terms from three vocabularies: ASM Handbook Volume 21 [ 9 ], MIL-HDBK-5 [ 14 ], and MIL-HDBK-17 [ 12 ]. The materials science community was in need of a system that could: (1) consolidate these vocabularies, (2) represent each dictionary term and its metadata such as, provenance appropriately, and (3) provide the community broader and easier access to materials science terms and de nition.

These requirements which were presented to us by AFRL are ful lled via our KnowledgeWiki tool. We present its development for ful lling the requirements along with some preliminary results in the rest of the paper. 3.

The Semantic Web community leveraged crowd-sourcing techniques for the purpose of data curation [ 15 ]. Semantic wikis, which enhance traditional wikis with structured knowledge representation, have been used to facilitate collaborative development of ontologies and knowledge bases [ 3 ]. Semantic MediaWiki [ 10 ] is a widely known semantic wiki and it has been used to capture semantic data in variety 1https://www.w3.org/2009/08/skos-reference/skos.html 2https://www.w3.org/TR/prov-dc/ 3http://qudt.org/ of areas including healthcare [ 2 ], energy, media, and air navigation. Semantic Mediawiki is currently being used in over 300 public wikis around the globe. Some of the other examples of Semantic wikis include: OntoWiki [ 1 ], IkeWiki [ 16 ], SweetWiki [ 4 ], and Acewiki [ 11 ]. More details on di erent kinds of wikis can be found in [ 3 ]. However, none of these wikis have a simple way to incorporate metadata into their data models, a challenge our work addresses.

As mentioned earlier Wikidata is the only crowd-sourced application which provides the capability to incorporate metadata. Wikidata [ 18 ] is a prominent community-oriented effort to create and curate an online, structured knowledge base that can be used by every language version of Wikipedia. For example, the de nition of a term can be further accompanied by its source and license information. In order to represent this in RDF, Wikidata uses auxiliary node in a way that is similar to a blank node. Figure 1 shows an example of the modeling with an auxiliary node. Here, property P26 (De nition Text) is broken into two properties to use the auxiliary node to link to the context information. As they mentioned in their paper, this led to Wikidata properties not directly corresponding to properties in RDF.

The widely-known techniques for incorporating metadata into the RDF data model are: (1) rei cation, (2) n-ary relation, (3) the singleton property, and (4) a named graph. In the recent work by Hernandez et al., [ 8 ] the authors compared these techniques for reifying RDF triples, with the goal of representing Wikidata as RDF, which would allow legacy Semantic Web languages, techniques, and tools to be used for Wikidata. They reported that the singleton property approach o ered the most concise representation on the triple level. Similarly, this approach also o ers the most compact dataset according to the experimental comparison in the PubChem dataset [ 7 ]. Therefore, we adopted the singleton property approach [ 13 ], which was originally developed by our group, to represent the metadata about the RDF triples.

Existing work, including Wikidata, lacks support for importing an existing RDF data set with metadata. Even though the SMW extension RDFIO4 provides the capability to import arbitrary RDF triples into the wiki, it does not have the capability to handle an RDF data set with metadata information. The proposed RDF import extension automatically identi es the RDF subgraph structure of the dataset and imports the dataset while preserving the 4https://www.mediawiki.org/wiki/Extension:RDFIO

In this section, we rst describe the overall architecture of the system. Next, we describe how we developed the new extension with singleton property templates for representing the provenance metadata of the triples. Then, we describe the algorithm for identifying the singleton property templates for any given RDF dataset. The overall architecture of KnowledgeWiki is shown in Figure 3. 4.1

The following sections describes the architecture overview of our system (see Figure 3). We rst explain how the data is collected via the existing semantic forms and how the singleton property template is integrated into the SMW. Next, we describe our new data representation module for SMW. We also describe how each entity is processed and, how the triples are created and, how each entity is represented on a wiki page. Finally, we discuss how more complex operations, such as CSV data import and RDF import/export are performed.

Data Collection. Templates are an integral part of Semantic MediaWiki and the simplest way to give input. Each template is de ned as a set of eld and value pairs. Each eld in the template is mapped to a separate semantic property. Semantic properties in SMW are used to express binary relationships between semantic entities. The eld values can be set to a number of prede ned datatypes, such as Text, URL, Page, Boolean, Number, etc.

For example, a regular template which provides the text de nition for a given term is de ned as follows: {{Definition Text | Definition = }}

For the term Autoclave, the page would be titled Autoclave and this page would contain the regular template \De nition Text" along with the property name and its value as follows: {{Definition Text | Definition = A closed vessel for producing ..}}

As mentioned earlier in Section 4.1, templates are an integral part and the simplest way of including markup in the wiki. We introduced the singleton property template in Section 4.1. and in this section we describe the implementation details. The singleton property template is an extension that we implemented for representing the metadata of RDF triples such as provenance. Since we developed a new template for the purpose of disambiguation, we name the traditional SMW template as a regular template and our new template as a singleton property template. Every singleton property template contains a set of eld-value pairs: {{Singleton Property Template name | property_1 = value_1 | property_2 = value_2 | property_3 = value_3 }}

Each eld is mapped to a semantic property. Within each template, a property selected for instantiating singleton properties and is speci ed in the MagicWord parameters associated with this template. A sample de nition for the singleton property template is shown in Figure 2. We introduce the MagicWord \#singletontemplate" for processing the singleton property template, for example: #singletontemplate:singleton prop=property 1.

When creating the singleton property template, the user gets to select the property to instantiate the singleton property. Each template can contain only one singleton property. This singleton property will bear the metadata of the RDF triples and the semantic forms can contain more than one singleton or regular template. In this case, the singleton property will be created for the property \property 1." It is associated with other meta properties such as \property 2" and \property 3." The singleton property template above will be mapped to this set of triples as seen in Table 1.

For instance, for the term Autoclave, we have a term definition for it. This de nition has other metadata associated with it, such as the source of the de nition, which is provenance information. We have the rights or the license information associated with this de nition. As discussed earlier, the regular template doesn't have support for this kind of representation. With the singleton property template, we can represent the metadata requirement in this example. Particularly, we de ne the singleton property template for the term Autoclave as follows: {{Definition Text | Definition = A closed vessel.. | Source = ASM handbook Volume 21: Composites. | Rights = Reproduced with permission of ASM International. All rights reserved. www.asminternational.org}}

For processing this singleton property template, the MagicWord parameter is de ned as: #singletontemplate:singleton prop=De nition. This singleton property \De nition" will bear the metadata of the RDF triples, such as \Source" and \Rights." The singleton property template above will be mapped to this set of triples as seen in Table 2.

The advantage of our approach is that the singleton property template extension was seamlessly incorporated into the existing extensions. We will demonstrate the use of the singleton property extension for curating materials science vocabularies in Section 5. 4.3

As we extend Semantic MediaWiki to develop KnowledgeWiki for curating the vocabularies, we identify its potential as a curation platform for any given RDF dataset. To exploit the full potential of this crowd-sourcing platform for the curation of RDF datasets, importing the RDF dataset into the Semantic MediaWiki platform is required. Hence, we developed an extension to import an existing RDF dataset into Semantic MediaWiki.

We developed an algorithm to create pages in Semantic MediaWiki for each entity given the RDF dataset as an input. It also has the capability to handle the provenance information included in the dataset by reusing the singleton property template.

mv:Termi skos:Concept rdfs:SeeAlso

rdf:type rdf:type

mv:Term mv:synonym

vaem:abbreviation synonym text

In this work, we were particularly interested in understanding how our KnowledgeWiki could be utilized in materials science community. We would like to explore this question quantitatively and qualitatively with respect to the materials science domain. The architecture of our KnowledgeWiki was discussed in Section 4.1. Here we describe a speci c use case of the system for materials science. 8http://matvocab.org/wiki-dev/index.php/Special: Importdataset

Schema-level description of vocabularies. We were provided with three vocabularies by the community, including ASM Handbook Volume 21, MIL-HDBK-17, and MILHDBK-5. The data provided to us was legacy data present in excel spreadsheets. We worked with domain scientists and identi ed the following requirements: (1) consolidate these disparate sets of vocabularies to have one common vocabulary describing the materials science domain, (2) develop a method to represent each term and its provenance metadata appropriately, and (3) make this wiki-based tool open for community authoring.

Singleton Property Templates with semantic interface for crowd-sourcing. To address the requirements described above, we developed a new data model for representation. Our data model captures elements such as definition text, image, sound and other elements to represent each term. We de ned eleven templates to model and represent the materials science vocabulary, including six singleton property templates and ve regular templates. The singleton property templates are De nition Text, Image, Video, Sound Recording, Equation, and Code Snippet. The regular templates are Name Abbreviation Synonym, Symbol, Other Website De nition, Unit, and References.

We use existing vocabularies such as dcterms (Dublin Core), skos (SKOS), and qudt (Quantities, Units, Dimensions and Data Types) to de ne the classes and properties. We also create a new pre x mv (MatVocab) for the new integrated materials science vocabulary. Figure 4 shows the schema-level data model that we developed using the singleton property template extension. The property mv: TermskosDe nition001 is a singleton property instance of the imported property skos:de nition and is used to specify the de nition of a term. Using the singleton property data model, we can specify all the metadata associated with the de nition such as provenance information dcterms:source and rights information dcterms:rights. When creating a singleton property template, the user gets to select the property to instantiate the singleton property. For example, in Figure 4 the \skos:de nition" is selected to generate the singleton property.

The singleton property template uses the singleton property instance to refer to the entire triple succinctly, and enable metadata to be associated with triples via the property instance. Given a vocabulary term, our data model allows users to add multiple de nitions (with associated details) for any element. For example, the term Autoclave can have multiple de nitions derived from di erent vocabularies such as the ASM Handbook Volume 21, MIL-HDBK-5, and MILHDBK-17. Singleton property templates can handle such representations easily.

Finally as mentioned earlier, using these templates along with the forms a wiki page is created for each entity with the term name as the page title within the main namespace.

Speci cally, Table 4 describes a simple example where, matsup could be the namespace URI coined by a materials supplier coupled with a local part that speci es the materials. Likewise, testhouse could be the namespace URI devised by the company testing the materials (perhaps only used on their intranet). The existing pre x URI rdf and dctype include a number of de ned terms (local parts) that are widely used.

Loading vocabularies into KnowledgeWiki. The three vocabularies (ASM Handbook Volume 21, MIL-HDBK5, and MIL-HDBK-17) were imported into the wiki using the \CSV Import" feature9. The implementation described so far, allows users to add terms via semantic forms. However, in the case when data providers have a large number of terms, it is a tedious task for them to go and add each term manually via semantic forms. Therefore, we developed a bulk upload functionality to add a large number of terms to KnowledgeWiki using a prede ned structured form. We extended the Semantic MediaWiki Import CSV feature.

We restrict the format of the input CSV le in a way which adheres to our data model. More speci cally, we only allow the properties supported by our semantic forms. In the CSV le, the header row speci es the properties and other rows specify the values for each term (Title). A sample CSV le segment is shown in Table 3.

Here the user just needs to upload a CSV le from their le system. Once the le is uploaded, KnowledgeWiki represents each entity on a wiki page and generates the semantic triples. For example as seen in Table 3.

In order to help the materials science community to create a consolidated vocabulary (MatVocab) using our system, we had to address some complex problems when dealing with knowledge integration. For example, more than one de nition can be present on a term's page. The community can use KnowledgeWiki as a means to express member opinions and ultimately select speci c elements as terms to de ne. In other cases, what started as a de nition for one term 9http://matvocab.org/wiki-dev/index.php/Special:ImportCSV could evolve into multiple terms with their own respective de nition elements. For example, if \Modulus" were added as a term then, through community discussion that speci c term could be spun-o into other terms like TensileModulus, ShearModulus, CompressiveModulus, or BulkModulus.

One unique feature of KnowledgeWiki is the ability to allow for multiple textual de nitions along with their respective source and license information. Therefore, whenever an element of a de nition is used, the license is presented along with the element.

The three vocabularies were successfully imported into the wiki using our extensions. There is a total of 2,800 entity pages created from the three vocabularies. Each of these terms may have values for 11 templates de ned on the wiki. Currently Kno.e.sis is hosting an instance of KnowledgeWiki, called MatVocab, that is being used to create and curate a vocabulary for material scientists.

Open Sourcing. Open sourcing a piece of software is to make its source code available for modi cation or enhancement by anyone. By making the software open source, it becomes transparent, easy to use, and widely accepted in the community. It allows for communities to freely access and modify the software and customize it per their requirements. The implementation of our KnowledgeWiki is open source and free to use. The software is available on GitHub10. 6.

One of the important features of our wiki is how easily it can be adapted to a new domain, as discussed below.

1) Creating the data model for the vocabularies from other domains. In order to create a vocabulary for a new domain, we only need to create the set of templates to re ect the new relationships in the domain. For this, the user has to identify the information that the user wants to capture, and based on requirement, the user can use a singleton property template 10https://github.com/MaterialWays/semanticwiki/tree/ deployment/deployment/README.md

Subject matsup:Material 7075T6 testhouse:TestSpecimen 3 testhouse:TestSpecimen 3 or a regular template. For example, if the user wants to capture provenance information for an entity, the user should use the singleton property template. Next, the user needs to create the semantic forms using these templates. These forms are used for creating and curating information on the wiki. Further information on how to set up KnowledgeWiki for a new domain can be found on our wiki page11

This work can be applied to other domains such as chemistry (e.g., PubChem) and bioinformatics (e.g., BKR). In fact, the singleton property approach has been discussed by the bio hackathon community and has also been evaluated and compared with other approaches in PubChem. In another work by Tudorache et al. they describe their tool iCAT to curate the International Classi cation of Diseases (ICD-11) [ 17 ]. They report the use of such a tool in the classi cation of epidemiology and healthcare data management for clinical purposes. Therefore, with the above mentioned features we feel our system is easy to adapt to a new domain. Another aspect is the

2) Use of our system for creating RDF datasets in general, this is not only with provenance metadata, but for any kind of meta data about the triples. The semantic extension that we developed for representing meta information is generic and allows for any kind of data annotation such as provenance, access control or spatio-temporal information. Finally,

3)Data re-usability, the data and its meta information that is collected or created using our system is ultimately represented as triples in the RDF format. This output created by our system, where the data is represented using the singleton property approach can be queried or reused by other applications and tools. The singleton property triple pattern allows consumers of our data to discover expected access patterns for this data. Currently our data can be queried using the standard SPARQL queries, we have listed a set of simple SPARQL queries on our wiki page12.

In this work we have reported about the development of our KnowledgeWiki and how we adopted it for curating vocabularies in the materials science domain. We have showed that by enhancing and extending the existing Semantic MediaWiki we can facilitate communities to create and curate vocabularies. KnowledgeWiki has been designed as an extension to the well known open source platform Semantic MediaWiki (SMW) that adds a mechanism to capture provenance information e ciently by leveraging the well known singleton property approach.

KnowledgeWiki is open source and licensed under GNU General Public License. The content published in KnowledgeWiki is licensed under Creative Commons AttributionShareAlike License unless overridden by the data providers 11http://wiki.knoesis.org/index.php/KnowledgeWiki 12http://wiki.knoesis.org/index.php/KnowledgeWiki as stated in the new licensing statement. We have been able to bring existing materials science data into the wiki successfully for curation and preservation the metadata of the RDF triples. The extension of singleton property templates in KnowledgeWiki is able to accomplish such a goal. 8.

This research was funded by the U.S Air Force Research Laboratory(AFRL), through the contract FA8750-13-1-0244. We would like to thank Clare Paul for his vision of how domain experts would use the KnowledgeWiki as well as his technical contributions in its design and development. We also thank Kalpa Gunaratna, Siva Kumar Cheekula, Swapnil Soni, and Mary Panahiazar who contributed to related aspects of this work.