Information sharing can be effective with structured data. The Semantic Web is mainly aimed at structuring information by creating widely accepted ontologies. However, users have different preferences and evolving requirements. It is not practical to attempt perfect schema definitions with strict constraints. Creating structured formats should be a collaborative and evolutionary process. Social software motivates wide participation by providing easy interface. We propose a system called StYLiD for sharing a wide variety of structured information. Users freely define their own structured concepts. The system consolidates different versions defined by different users. The attributes of the different concept versions are aligned semi-automatically into a single unified view. Popular concepts gradually emerge from the concept cloud and stabilize. Concept definitions are flexible. An attribute value can take a literal or a resource URI and the suggestive range does not constrain the contributors. StYLiD generates unique dereferenceable URIs so that data items can form a linked data web. Structured data is embedded in machine readable form using RDFa. Search and browsing features are provided to utilize the structured data and consolidated concepts.
Structured data, information sharing, social semantic web, concept consolidation, collaboration, linked data, RDFa
Information sharing on the Web has become a basic need in communities. We want to share a wide variety of information. It would be desirable to have some system which can facilitate the modeling and sharing of such heterogeneous
• It becomes easy to define the semantics of data and make it machine understandable so that processing can be automated. • Information sharing becomes more effective when data is structured following common conventions. • Search and browsing becomes more effective with structured data. • Structured information can be easily mixed. It becomes easy to integrate information from various sources. • We can have interoperability between different systems by forming standard formats. Even multiple structure definitions for similar data may be mapped to each other.
Thus, structured data becomes open and shared for all
rather than being closed in proprietary systems. With the
growing significance of structured data, the Web is rapidly
moving towards a Structured Web which can be a
transitional step towards the Semantic Web and can be fully
realized with current technologies[
Efforts for the Semantic Web have been mainly being
directed towards creating standard formats in the form of
ontologies. However, currently there are not many ontologies
to cover the wide variety of information we may want to
share[
Creating new ontologies and information systems is not easy. Data modeling is a difficult task. It should be flexible to accommodate requirements and exceptions that surface in the future. Users may need different data and varying levels of details depending upon the purpose. Moreover, people have different views and should be allowed to maintain their preferences. It is not practical to impose a single standard or strict constraints.
Thus, creating ontologies or common formats should be
a widely collaborative process[
On the other hand, social software has proven to be successful in drawing huge user participation and contribution. Tagging is successful because it is very simple and anyone can contribute easily. Systems like tagging and social bookmarking do not impose any hard constraints for sharing data. However, these systems do not provide much semantic structure to information. Though some social software systems do provide structured data, they are closed systems with less interoperability and integration with other systems.
Recently, the combination of social software with
Semantic Web technology towards a Social Semantic Web has been
gaining significant attention[
We propose a system called StYLiD (an acronym for
Structure Your own Linked Data) which gives users the freedom
to define the structure of their own data. It is easier to
define one’s own quick data model than to search for suitable
ontology or schema and understand it. We propose to let
the users input information freely without imposing any
constraints, just like tagging. Computations can be done later
to consolidate similar concepts, deal with inconsistencies and
align multiple definitions. Concepts can gradually converge
to stability by usage in the same way as folksonomies. The
quality and stability of data is maintained when many
eyeballs are watching and people can vote contents. This has
been demonstrated well by social sites like Wikipedia1 and
Digg2. Furthermore, StYLiD is an open system that can link
to external data and allows others to link in for building a
linked data web[
We discuss some use case scenarios in Section 2. We describe the StYLiD platform in detail in Section 3. Section 4 gives some details about implementation. We discuss some related works in Section 5. Finally, we conclude in Section 6 and state some ongoing and future work.
Suppose a user wants to share some structured information. However, he cannot find a suitable schema or any system for handling such data. He may freely register an account on StYLiD and define his own structured concept on the fly. He simply enters the concept name and a list of attributes. If a similar concept already exists in the system, he may choose to use the concept directly and enter instances or modify the existing concept to create his own version. He may modify his own concepts later and add more attributes whenever needed.
The user may easily start sharing data using his own structure definition. Any other registered user can also use his concept and contribute instance data. While entering data the system helps the user by suggesting range of values for the attributes. The user may easily pick instances from this range. However, any suitable value may be entered even though it is not in the suggested range. The user may easily type in literal values for attributes. If the user knows a resource URI for the value, it may be entered to link to that resource. The corresponding resource may also be entered later and the original entry edited to specify the URI link. All the user defined concepts are visualized as a concept cloud where popular concepts are seen bigger. The user can browse different types of data with the concept cloud. When the user hovers over any concept, the attributes and description of the concept are shown so that the concept and its structure can be instantly understood (see Fig.5). This is useful to see how well defined the concept is and whether it is appropriate for him. He may wish to view only the concepts defined by him or any particular user as a concept cloud. He also maintains a personal concept collection of useful concepts, also viewed as a concept cloud. Instances of a concept can be viewed in a record view or a table view. The user may switch between these views. The user can navigate through linked data entries. The data entries may also link to external resources. The user may search data instances using a simple web-based interface by specifying the concept name and a set of attribute name, value pairs as criteria. Advanced users may directly query the system using a SPARQL query interface.
Different versions of a concept defined by different users are consolidated by the system and shown as a single virtual concept. The different versions are grouped together in the concept cloud. The individual concepts in a group can be identified by visible labels for the creator name and version number. By clicking on a consolidated concept, the user would be able to see all the instances of all versions. He may want to see all the instances of a concept regardless of the creator or version. He may also want to see all the instances of a concept defined by a particular user regardless of the version. He may want to see only the instances of a particular version defined by a particular user. The consolidated concept cloud offers the desired granularity.
When the concept is a single distinct concept, the table view is straightforward, each attribute displayed as a column. However, when it is a consolidated concept, the corresponding attributes of the individual constituent concepts have to be aligned first. The system automatically suggests alignments in a form-based interface. The user may update this and add mappings not suggested by the system. Then all the data can be viewed in a unified uniform table view. The user may also rename the attributes of the integrated view and hide unwanted columns if needed to get a customized view. 2.5
The system embeds machine understandable RDFa3 data
in the HTML posts. An RDFa aware browser would be able
detect such contents and offer suitable operations for the
user. Many RDFa tools and plug-ins are becoming available4
and we may expect more powerful tools to be available in
the future. The use of RDFa has also been demonstrated by
recent works on semantic clipboard[
The StYLiD platform realizes the use cases described above. It enables the users to define their own concepts on the fly and share structured data. The main contributions of the system are as follows. Details are provided in the subsections to follow.
• Sharing structured data with user-defined concepts. Users may define their own concepts with attributes, freely and easily, and share structured data using them. Different users are allowed to have different versions of the same concept. Users can share, reuse and refine such concept definitions. • Consolidation of user-defined concepts. Multiple versions of concepts defined by different users are consolidated and corresponding attributes are aligned to produce a unified consolidated view. Popular concepts emerge out from the cloud of concepts. • Flexible definitions and relaxed data entry. Users are allowed to input information freely, according to their needs and preferences, instead of attempting perfect schema definitions and imposing strict constraints. 3http://www.w3.org/TR/xhtml-rdfa-primer/ 4http://rdfa.info/2007/02/12/call-for-proposals-rdfa-utilsservices/ • Open system for creating linked data. The system allows open access to its data using open standards. It can link both internal and external data to support a linked data web.
StYLiD is still a prototype and development is going on. A demo installation is available online5. Currently we are populating some sample data in the academic domain with different versions of concepts like faculty, courses, seminars, etc. Heterogeneity is common in such data because academic institutes have different systems and formats. Most of the data is being populated with the help of scrapers created using the free online service, Dapper6. We intend to continue using StYLiD in this domain with real users. However, the system can be installed and used for any other domain or general purpose. 3.1
The main interface of StYLiD is shown in Fig.2. The users of the system may freely define their own concepts by specifying the concept name, some description (optional) and a set of attributes. Each attribute is defined by the attribute name, description (optional) and a set of concepts as the suggested value range (optional) as shown in Fig.3. Any user may enter instance data for the concepts using the interface shown in Fig.4. An attribute of a concept can take a single value or multiple-values. Each of the values may be a literal or a resource (identified by its URI). If the value is a resource URI, a human readable label may be entered along with the URI.
The system allows different users to define their own concepts having the same name. Moreover, users do not need to define concepts from scratch. The user can modify an existing concept to make own version. However, users are not allowed to tamper with others’ concept definitions. The system makes a copy of the concept and allows the user to make modifications on it. It keeps record of the source from which the modified concept was derived using the dc:source property. Users can update their own concept definitions keeping the existing instances consistent. Attributes can be
6http://www.dapper.net/ added. However, if we need to rename or delete attributes of the concept a new version of the concept should be defined to keep the existing data intact. Thus, the same user can also have different versions of his/her concept with the same name.
Structured Data Formats. The system embeds machine readable structured data in HTML using RDFa format. It also outputs the data in RDF format separately. Thus, the system produces formal machine understandable contents though the user interface is quite simple and informal like a tagging system.
A Personal Structured Data Space. The system offers every user a personal structured data space. It provides a Concept Collection for each user, as seen in Fig.5. Concepts created or adapted by the user are automatically added to this collection. Besides these, users can also add any other useful concepts to their collection. The users need not be overwhelmed by the huge cloud of concepts defined by the large number of users. Moreover, the concept collection is also helpful to mark the concepts that the user has been using out of numerous concepts and different versions. The concepts actually created by the user are also shown in a separate tab. 3.2
Creating perfect concept definitions with strict constraints is not easy and practical. It is difficult to think of all attributes and all possible value ranges at the time of concept definition. It may also be difficult to say whether an attribute value would be a literal or a resource and whether the attribute would have a single value or multiple values. While defining a concept A, if an attribute must take a resource of type concept B, we must first ensure that concept B has already been defined. If concept B has an attribute which takes resource values of type concept C, then concept C must be defined first, and so on.
Similarly, at the time of instance data entry, it may be difficult for the user to enter perfect data as mandated by a schema. All attribute values may not be known. Proper resource URIs for attribute values may not exist or the user may not be able to find it at the time. Moreover, exceptions may always exist no matter how well the schema has been designed and unpredicted new data instances may appear.
The system tries to avoid these difficulties in data modeling and data entry by allowing flexible and relaxed definitions. The concept definition may be incrementally updated later and new attributes may be added. New versions of the concept may be defined by different users or even the same user. The range of values defined for attributes, as seen in Fig.3 and 4, is only suggestive and do not impose strict constraints. Rather the system assists the user to fill data using the suggested range. The suggestive range may be updated later by including more concepts or narrowing down to refine the range. The system accepts literal values though resource values may be desirable for an attribute. Instances may be updated later to change a literal into a resource value by adding the URI. The users may input single or multiple values for any attribute as appropriate. With such relaxed data entry interface, of course, we may get some imperfect, incomplete or heterogeneous data. However, users generally enter appropriate or sensible data for their purpose. This has been evidenced by systems like tagging and wiki which accumulate large volume of good data in spite of having completely relaxed interface. 3.3
Concepts defined by different users with the same name are grouped together by the system. This forms a single virtual concept which consolidates all the grouped concepts. This consolidated concept can be used to retrieve all the instances though different users have different definitions for the concept name.
If C1, C2, ...Cn are the concepts defined by users 1, 2,... n with concept name “C”, the consolidated concept is given by C = C1 ∪ C2 ∪ ... ∪ Cn
Further, different versions of the same concept defined by a single user are also grouped together. Thus, we can obtain all the instances of a concept defined by a user irrespective of the version.
If Ci1, Ci2, ...Cim are the versions 1, 2,... m of concept “C” defined by the user i, then the consolidated concept for the user is given by Ci = Ci1 ∪ Ci2 ∪ ... ∪ Cim 3.3.1
Consolidated Concept Cloud
All the concepts contributed by different users are visualized together as a Concept Cloud, similar to a tag cloud. Better concept definitions will satisfy more users and will have more instances. Popularity of concepts is visually highlighted by increasing size. Popular concepts will receive more attention and motivate more use in turn. Thus, stable definitions will gradually emerge out from the vast cloud of concepts as more instance data are contributed. Clicking on any concept shows all instances of the concept. ply based on the Levenshtein edit distance7 between the attribute labels. So slight variations on spelling and morphology are easily handled.
It is not possible to make the alignment fully automatic and accurate. Moreover, alignments may vary for different users and for different purposes. So it is desirable to have the user in loop though the system greatly simplifies the work by providing automatic suggestions. The user can complete the process by adding matching attributes that the system could not detect or modify the suggested mappings. Thus, we propose to use both machine intelligence and human intelligence for the alignment process.
A Unified View. Each set of aligned attributes can be considered as a single consolidated attribute for the consolidated concept. The system automatically fills a name for each consolidated attribute, as shown in Fig.6, though the user may rename it as desired. The user may even remove attributes from the unified view, if not required. Thus, the user can create a unified view of the consolidated concepts, customized according to his need, and view heterogeneous instance data in a uniform table.
A consolidated concept formed by grouping different versions can be expanded into a sub-cloud. The sub-cloud shows all the versions of the concept defined by different users, labeled with the user name. Further, in the sub-cloud, if multiple versions are defined by the same user, they are subgrouped together. In the Fig.5, the “Faculty” concept has been expanded to show two versions by the user “god” and one version by “aman”. The sizes of all the different versions in the sub-cloud add up to form the size of the consolidated concept. Clicking on the consolidated concept shows all instances of all the versions of the concept. Similarly, we can also see all instances of the multiple versions of a concept defined by a single user by clicking on the user name. 3.3.2
Semi-Automatic Concept Alignment and Unification
Different concepts in a consolidated group are aligned to produce a uniform and integrated view. When the instances of a consolidated group of concepts are viewed as a table, as shown in Fig.7, the system automatically suggests alignments between the attributes of the concepts, as shown in Fig.6. Matching attributes are automatically selected in the form-based interface. Currently, the mapping is sim
version number (if the same user has defined different versions of the concept).
An example URI for a concept “Car”, version 2, defined by the user with ID 1 would be like http://www.stylid.org/stylid/concept detail.php? concept name=Car ver2 1#car
Similarly, consolidated virtual concepts are also assigned URIs so that they can be uniquely referenced. An attribute is uniquely identified by the concept and the attribute name.
For example the URI for the price attribute of the car concept would be http://www.stylid.org/stylid/concept detail.php? concept name=Car ver2 1#price
An instance is uniquely identified by the system generated ID for the instance. The URI of an instance is different from the URL of the post showing it. A concept URI dereferences to a page describing the details. An instance URI dereferences to the post showing its details. The details page contains both human readable and machine readable data.
Data instances can be linked to each other by entering resource URIs as attribute values(see Fig.4). The linked data is manifested as simple hyperlinked entries for the user (see Fig.2). However, the linking of URIs helps in the creation of a linked data web, not just hyperlinked pages. The system can link to URIs from any system on the Web. On the other hand, it allows others to link in to its data by providing unique dereferenceable URIs.
StYLiD is an open system that does not lock data into itself. Besides allowing others to link in, the system facilitates the reuse of structured data. Structured information snippets in embedded formats like RDFa may be posted elsewhere or distributed via social media. The system provides an advanced search interface, as shown in Fig.8, which can be used to retrieve instances of a concept specifying attribute, value pairs as criteria. The system also provides a SPARQL query interface for open external access.
Fig.9 shows the system architecture of StYLiD. It is built upon a social software platform for harnessing user contributions. The social software provides all the basic features such as content management, assessing popularity of contents, user management, social networking and communication among users. The concept management component enables the users to define their own structured concepts. The component handles the different versions of concepts defined by different users. The structured data management component gathers the instance data contributions from users. The concept management component also handles URI management by assigning each of the concepts and instances a unique dereferenceable URI. The system links structured data items using the URIs. The concept consolidation component consolidates multiple versions of a concept defined by several users. It maps the different versions by aligning attributes and provides a unified interface for the consolidated concept. The structured data embedding component embeds structured data in HTML output using RDFa. RDFa is W3C supported and a comparison with other embedded formats8 indicates that it is a reasonable choice. The system produces snippets with embedded structured data which can be posted elsewhere. All the concepts and structured data contributed by users are stored in the collaborative data store coupled with the social software. The structured concepts and data are stored as RDF triples in a MySQL database. The system provides some services to exploit the structured data like structured browsing, search and query and allows RDFa driven features discussed in the use case scenario (Section 2.5).
StYLiD was built upon Pligg9, a popular Web 2.0 content management system. This open source social software has a long list of useful features and strong community support and, furthermore, provides extensibility. It uses PHP and MySQL. We used the RDF API for PHP (RAP) as a Semantic Web framework to manage structured data. 5.
There have been several recent works on collaborative creation and sharing of structured data on the web. Freebase10 is one of the most prominent works. Similar to Google Base11, it allows users to freely define their own structured types and input instance data. However, Freebase keeps the structured types defined by different users separate. It does not consolidate or relate similar concepts. Even concepts having the same name are not shown in a combined way. User defined types and domains are kept within the user’s personal space and not easily promoted to the standard types and domain collection. So it is difficult to leverage the structured concepts defined by the large number of users. Moreover, it is difficult for casual users to create their own 8http://bnode.org/blog/2007/02/12/comparison-ofmicroformats-erdf-and-rdfa 9http://www.pligg.com/ 10http://www.freebase.com/ 11http://base.google.com/ types in Freebase because of the strict constraint requirements. All the attributes must have strict types and the range should be within the types already defined in the system. The attribute and range definitions cannot be altered later if some instances of the concept already exist. Further, it may also be difficult to enter instance data in Freebase because of strict schema constraints. If an attribute takes as value a resource of some type, the resource must be entered first. Although Freebase has made a lot of instance data available by scraping data from vast sources like Wikipedia and MusicBrainz, a non-existing instance must be modeled and entered by the user. Freebase interlinks instance data to each other as attribute values. However, it cannot link to external resources at the data level and it is difficult for other systems to link to Freebase data resources.
The myOntology[
There had been various works on semantic blogging[
Works have been done on deriving ontologies from
folksonomies[
Revyu[
Exhibit[
We proposed StYLiD as a single platform for sharing a wide variety of structured data. Users can freely define their own concepts. Relaxing constraints would encourage more user contribution to better meet their requirements. The task of consolidating, aligning and unifying user defined concepts can be handled by the system without bothering the users much. Although several definitions of a concept may exist, the system can provide a single consolidated view so that even heterogeneous structured data can be handled uniformly. It also facilitates the emergence of popular and stable generalized definitions. Keeping the system open and adopting URI conventions support the creation of a linked data web. Thus, even with the informal base of social software we may produce formal machine understandable structured data which can be shared, interlinked and integrated.
In the future, sophisticated schema mapping techniques[