The Semantic Web

Linked Data

Tim specifies four qualities necessary for data to be considered linked:

1. "Use URIs as names for things.

2. Use HTTP URIs so that people can look up those names.

3. When someone looks up a URI, provide useful information using the standards

4. Include links to other URIs. so that they can discover more things." [12]

This definition of Linked Data specifies the conditions necessary for Semantic Web-supporting data. The first condition guarantees names on the Semantic Web, just as on the Web, dereference uniquely, regardless of whether or not they indicate a resource. However, it also expands the types of things URIs can represent beyond Web documents; on the Semantic Web, a URI can refer to any kind of information, whether it be a person, a document, an application, or a value. The second condition applies to content on the Web and on the Semantic Web equally, and ensures the Semantic Web is compatible with the World Wide Web.

The final two conditions for Linked Data are what differentiate the Semantic Web from the Web. A Semantic Web URI dereference must by default provide the dereferencer with both standardized information about the dereferenced URI (condition three) and a set of URIs related to the dereferenced URI (condition four). A dereference on the Semantic Web does not necessarily link to a document; instead, it provides information about that URI and how it is connected to other content on the Semantic Web. These characteristics make Linked Data self-describing.

Queries

Linked Data is not a goal in and of itself; rather, it is a prerequisite for query-resolution by machines, the core feature of the Semantic Web. Query-resolution upon Linked Data is significantly different from Query-resolution in traditional relational databases: over Linked Data, a result is obtained by recursively exploring connections in an instance of Linked Data to other distributed instances of Linked Data; over relational databases, queries are assessed by accessing tables within that database. Relational databases are 'flat' and can be thought of linearly; Linked Data forms a graph [13].

Query-resolution over Linked Data has a number of advantages over Relational Database query-resolution, particularly in a decentralized and distributed environment like the Web. First, querying data spread across multiple sources is handled easily by the graph-structure of Linked Data, as opposed to requiring joining different relational databases. Second, because it Linked Data query-resolution occurs over HTTP, caching can be implemented easily [14].

Ontologies

The set of rules that define the validity of a statement and the logic of the inference rules for some Linked Data is called an Ontology or a Vocabulary [15]. Multiple Vocabularies can exist on the Semantic Web, though the value of a Vocabulary is closely related to the amount of information within it. Per Metcalfe's Law, a single large Vocabulary provides more value than two Vocabularies half the size [16].

An Ontology must be internally consistent to be useful, but there is no requirement that different ontologies be consistent between each other. This issue is discuess further in the Implementation Ontologies section.

Knowledge Representation

The Semantic Web differs from the Web by attaching meaning to content. The meaning of this content is used to facilitate machine-exploration and query-resolution. A Knowledge Representation system sets out a method to associate attributes with objects. Given the distributed nature of the Semantic Web, several special considerations must be taken in structuring a knowledge representation system.

• Vagueness

  Vague statements pose an interesting challenge for any knowledge representation system, and the way these statements are handled has a deep effect on what that system can be used for. A statement is vague when "there are possible states of things concerning which it is intrinsically uncertain whether [those states are] excluded or allowed by the proposition... because the speaker's habits of language were indeterminate" [17]. The sentence "Hunter is young" is a vague sentence, as the meaning of 'young' is contenxt dependent.

Knowledge Representation systems that are able to describe an unbounded number of subjects, objects, and relations are realistically guaranteed to contain vague statements. On the other hand, a bounded Knowledge Representation system might avoid the issue of vague statements by sufficiently restricting the set of allowable statements.

Inference

Inference is the last component of the Semantic Web necessary for automatic Query-resolution. In the Semantic Web, inference is the process by which "automatic procedures can generate new relationships based on the data and based on some additional information in the form of a vocabulary" [18]. There are three issues related to Inference which require special attention:

• Inconsistency

  Any Inference rule for use on the Semantic Web must deal with inconsistencies. This is a deep and fundamental logical issue, as in Aristotelian Logic, a contradiction entails anything, a result that is unacceptable on the Semantic Web [19]. A logical system is described as 'explosive' if, in that system, "A and not A entails B, for every A and B" [20]. Deafeasible Reasoning Systems, like paraconsistent logic, are non-explosive, promoting a weaker for of logic in which contradictions are permissible [21]. These kinds of non-explosive inference systems are generally used on the Semantic Web.

• Uncertainty

  Regardless of the Knowledge Representation System, there are some queries for which there is more than one possible answer. As an example, if the data is 'a polygon with four sides of equal length meeting at right angles' and the query is 'what shape is this polygon?', there are two equally valid answers, though they differ in specificity: 'square' or 'rectangle'. Probabilistic analysis can be used to determine the most likely answer in these situations [22].

• Vagueness

  If vague statements are allowed in the Knowledge Representation system, the inference rules must have some means of assessing their truth-value in order. Fuzzy Logic, where truth statements are assigned a decimal value between 0 and 1 (as opposed to a binary one in traditional logic), can be applied in this situation [23].

Knowledge Representation Implementations

• RDFa (Resource Description Framework in Attributes)

  The W3C proposes RDFa as the standard format for Semantic Web knowledge representation, with the aim of solving "the problem of marking up machine-readable data in HTML documents" [24]. RDFa uses subject-predicate-object triples, is capable of describing arbitrary relations and IRIs [25]. The subject identifies what the statement is about; the predicate identifies the type of relation in the statement; and the object indicates the value that the subject takes for the predicate. The W3C spec specifies that subjects and predicates must be IRIs, but objects may be literals. This requirement stipulates that subjects and predicates be unique references, while the object can have any value.

• Microformats

  Microformats are an alternative to RDFa. The key difference between microformats and RDFa is that microformats, unlike RDFa, are not "Infinitely extensible and open-ended" [26]. Microformats are meant to reuse existing data formats, like hCard, for Semantic Web markup and have a low barrier to entry.

• Microdata

  Microdata is another alternative, and supported by many of the major Search engines [27]. This format defines relations on the schema.org website. Like microformats, the microdata format does not allow arbitrary attributes to be defined. However, unlike microformats, which aims to reuse existing formats, the creators of microdata define their own attributes and relations.

A more complete overview on the differences between RDFa, Microdata, and Microformats can be found here.

W3C Ontology Implementations

The W3C has developed a number of ontologies, each for a different purpose.

• RDFS (Resource Description Framework Schema)

  RDFS is an ontology that extends the RDF Knowledge Representation Format, providing a "mechanisms for describing groups of related resources and the relationships between these resources." [28]

• OWL (Web Ontology Language)

  OWL, unlike RDF, RDFa, or RDFS, which define abstract ways objects can relate, is an ontology with a "formally defined meaning" [29]. This ontology is specifically intended to be used for information that "needs to be processessed, as opposed to situations where the content only needs to be presented to humans" [30].

• SKOS (Simple Knowledge Organization System)

  SKOS is an ontology which "aims to provide a bridge between different communities of practice within the library and information sciences involved in the design and application of knowledge organization systems" [31].

Other Technologies in the W3C Semantic Web Stack

There are a number of other tools in the W3C Semantic Web Stack.

• SPARQL (Simple Protocal and RDF Query Language)

  SPARQL is the W3C standard Semantic Query Language. SPARQL supports "aggregation, subqueries, negation, creating value expressions, extensible value testing, and constraining queries by source RDF graph" [32].

• RIF (Rule Interchance Format

  RIF is a W3C initiative which "aims to develop a standard for exchanging rules among disparate systems, especially on the Semantic Web" [33]. The initiative aims to develop a group of interoperable languages, using those language's rigorously defined semantics and syntax to accomplish this task.

Quantity

The massive volumne of pages and data on the Web is the single greatest challenge to the spread of the Semantic Web. There are 4.54 billion pages as of May 2015, and marking up each of those pages with Semantic Web metadata is far from trivial [34]. That said, the size of the Web is a double-edged sword, increasing the value of a widespread Semantic Web.

Barriers to Entry

The high learning curve of most Semantic Web technologies makes widespread adoption difficult. Compare the learning curve of HTML to the Semantic Web: to learn basic HTML, a user need only learn a few dozen tags; to learn 'basic' Semantic Markup, the user must have knowledge of how statements are written in the Knowledge Representation format, be familiar with the ontology they are working with right now.

Complaints about the high barriers to entry are levied against the W3C Semantic Web Stack in particular, and one of the reasons some choose to markup their pages with the simpler Microdata or Microformat Knowledge Representation formats. Ultimately, the decision to implement Semantic Content is a value judgement of the user. Given the endgoal of a global web of data, the ease with which content can be added to the Semantic Web is serious point.

Implementing Semantic Web markup does place a burden on the user, but this is a tradeoff for the gain in value that markup information provides. There's no free lunch - in order to get more out of a system, you have to put more in. The trick is to strike the right balance between expressiveness and ease of implementation.

Deception

Intentionally incorrect or misleading data poses a significant problem for the Semantic Web. The Semantic Web's key offering - increased convenience and access to data via automatic machine-exploration - is dependent on total user trust. As a result, the Semantic Web's margin for error when it comes to deceit is quite small, as even a small mistake can undermine the user's confidence in the System. Cryptographic techniques can be used to address the issue of authorship of a Semantic Web statement.

Censorship

The Semantic Web would make machine-understanding easier and more widespread, and it is possible this could be used to more aggressively censor content. I think this is one of the strongest points against the Semantic Web, as it presents a real potential negative effect of implementing such a global system. However, many of the arguments that advocate against its implementation could be made against the Web itself. Many powerful systems that provide real and significant value also open up new avenues for malicious actions; any tool can be used for good or bad. That said, this is a serious issue, and adequate provisions protecting the freedom of information need to be laid out. As the amount of content on the Semantic Web grows, this issue will become more serious, as it too obeys Metcalfe's Law.