Cognitive Computing - Semantically Speaking

[Note from the author: this blog is a reference to a series of blogs on Cognitive Enterprise Information Framework.  The order of the series starts with:

1. How Do I Describe My Information So Others Can Understand It?
2. Linked Data - The Foundation for Interchanging of Information
3. Linking Unstructured Data
   
4. Implementing a Cognitive Enterprise Information Framework (not yet published)

]

Cognitive Computing is many things to many people. The term cognitive  computing commonly implies that a computer can "possess the mental ability of knowing, including aspects such as awareness, perception, reasoning, and judgment." In this blog, the term cognitive computing is used in a narrower view for the purposes of defining a cognitive  enterprise Information framework that is based upon open standards for semantic understanding.

Semantic open standards are defined under the W3C Semantic Web Standards which divides the standards into two concerns:

• Syntax – Semantic Web (Linked Data)
• Semantic – Semantic Modeling (Knowledge Domain)

The Semantic Web was created by Tim Berners-Lee, the founder of World Wide Web and represents the form standard of Semantic Technology specifications published by the W3C. The Semantic Web represents the "Web of Linked Data" vs. as today, a "Web of Linked Documents" which is represented by HTML. The Semantic Web is the foundation of defining data structures, this does not in itself describe the semantics, or meaning behind the data.

Semantic Modeling, also addressed by the Semantic Web Standards, which allows you to define contextual relationships behind a well known vocabulary. It is the cornerstone of defining a knowledge domain. Semantic Modeling formalizes data that is contextually consistent across domains of knowledge. Semantic Modeling maintains fidelity between disparate data models, and provides the ability to describe a concept in one context to a concept in another context. Semantic modeling develops a picture that is greater than the sum of its parts and makes an enterprise system more robust, scaleable and agile.

The Semantic Web Standards is made up of a palette of technologies. It can seem a bit overwhelming if not presented in context. See below illustrations (Figures 1-4), to understand how to achieve a "Cognitive" Information framework.

Figure 1 - Web Enabling (Linked Document)

An explanation of graph shown in Figure 1,2, & 4 is in order. Starting on the lower left hand side, the graph illustrates where we are today with regard to knowledge representation capabilities.  The Web Enabling region of "Linked Documents" doesn't provide a very effective means for knowledge representation and is designed for content formatting. HTML represents a very inefficient way to represent knowledge and is costly to operate and maintain. It does not provide machine-to-machine interchangeability of knowledge.

The next step up for knowledge representation is Semantic Web Enabling (See Figure 2). This region introduces a new set of standards that provides the foundation for knowledge representation at the syntax layer.  This region provides the capabilities for "Link Data". The term Linked Data refers to a set of best practices for publishing and interlinking structured data on the Web. These best practices were introduced by Tim Berners-Lee in his Web architecture note Linked Data and have become known as the Linked Data principles. These principles are the following:

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 (RDF, SPARQL).
4. Include links to other URIs, so that they can discover more things.

Figure 2 - Semantic Web Enabling (Linked Data)

The benefits of this region is that it defines graph data through open standards, which is lacking in the No-SQL space. This region introduces a new concept for information representation called Triples. Triples are made up of three parts: Subject, Predicate & Object (see Figure 3 below for more details on Triples).

Below are a list of specifications and their definition that are illustrated in the Semantic Web region in Figure 2.

• RDF - Resource Description Framework is a highly canonical data model (in XML) and is the basis to represent any data form, any schema or conceptual structure. The RDF data is stored as a graph and the storage is called a “triplestore”, (a type of No-SQL database). Unlike a relational database, a triplestore is optimized for the storage and retrieval of triples.
• RDFa - Resource Description Framework in attributes is a W3C Recommendation that defines a set of extra attribute level extensions to XHTML for embedding rich metadata within Web documents.
• GRDDL - Gleaning Resource Descriptions from Dialects of Languages is a W3C Recommendation, enables transformation of XML documents, including XHTML to RDF triples.
• R2RML- RDB to RDF Mapping is a W3C Recommendation language for expressing customized mappings from relational databases to RDF datasets. R2RML mappers are themselves RDF graphs (R2RML Instance)
• RIF – Rule Interchange Format is a W3C Recommendation for exchanging rules among rule systems. RIF defines several “dialects” of languages and is not bound to RDF.
• SPARQL (pronounced "sparkle”) is a W3C Recommendation and is a recursive acronym that stands for: "SPARQL Protocol And RDF Query Language." SPARQL using SQL like syntax that allows users to execute queries against RDF datasets (triplestores). I will go into depth in a separate post on the benefits of SPARQL.

As mentioned above, triples is a statement about a relationship between two entities that are linked. Each triple has three parts:

• Subject
• Predicate (also knows as property)
• Object (Literal->Strings, Integers, Booleans or URI)

Each triple can connect to each other creating a set of Triples. A set of Triples are called a RDF Graph. The graph is stored in an No-SQL database called a Triple Store. See below Figure 3 for a graph example.

Figure 3 RDF Graph

The final step for knowledge representation is Semantic Modeling, the Knowledge Domain region - See Figure 4. In this region, true knowledge can be realized through an expressive framework for semantics-enabled data representation and knowledge inferencing. This region represents the highest effectiveness of knowledge acquisition, the lowest cost of operations, and is a highly desirable sweet spot to achieve for having a competitive advantage!

Figure 4 Semantic Modeling (Knowledge Domain)

The Semantic Layer provides specifications for defining a formal ontology. In computer science, an formal ontology is defined as: "a specification of a conceptualization."

Ontology is the modeling language for Knowledge representation and the underlying framework for realizing inferencing capabilities. Inferencing is the ability to start with a set of assertions. Example: Blue Jay is a bird, All birds have wings.  Does a Blue Jay have wings? - Yes (because we can infer that Blue Jays have wings since Blue Jay is a bird, and ALL birds have wings).

Inferencing capabilities are realized through an inference engine called reasoner. A reasoner is a piece of software able to infer relationships from a set of asserted facts (axioms) defined in an ontology.

Below are a list of specifications and their definition that make up the Semantic Layer.

• RDF Schema is a W3C Recommendation and is a semantic extension of RDF that supports a limited set of vocabulary. Whereas OWL, supports a full feature set of vocabulary that allows a more expressive ontology.
• OWL is a W3C Recommendation. Web Ontology Language (OWL) is a Semantic Web language designed to add an expressive framework for semantics-enabled data representation and knowledge inference.  There are 3 variants of OWL:
  • OWL Lite – Not Widely Used
    OWL DL – Descriptive Language base of First Order Logic for reasoning (which is most commonly used)
    OWL Full – Allows an ontology to augment the meaning of the pre-defined vocabulary. It is unlikely that any reasoning software will be able to support complete reasoning for OWL Full.

To better understand the differences between RDF, RDF Schema, and OWL, we will take a look at each of these specifications.

• RDF is an open standard for modeling information, were data is linked together with a predicate (a property defined as a verb).
• RDF Schema is an extension of RDF and defines a vocabulary of types (reserved key words) to describe the resources.  (Ex: <rdfs:Class rdf:ID="Person"/>).  RDF Schema allows you to type triples. If your familiar with Java, think Generics!
• OWL is an semantic extension of RDF Schema. OWL allows you to go further and define axioms about a relationship using one of the many powerful features called Characteristics.  As an example of one characteristics that was added as part of OWL 2 is the ability to perform property chaining (hasParent -> hasBrother = hasUncle).

The ability to take a domain related primary relationship (like hasParent in the domain of Family) and build new relationships from a base is truly transformational for modeling information.

To understand how to apply this technology, see my series of blogs outlined at the beginning of this article.