Semantics and Ontology in IT Data Management : Semantics and Ontology in IT Data Management It all started with Aristotle. But that’s a story for another time…. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 1 - 3
Agenda : Agenda Part I
Basic Concepts
Controlled Vocabularies
Taxonomies
Thesauruses
Ontologies
Inference Engines
Semantics
Semantic Interoperability Part II
Ontologies and Databases: an Action Plan
Develop a rigorous semantics.
Develop a formalized semantics.
Develop an integrated semantics.
Develop a late-bound semantics. Part III
Discussion Part IV
Final Thoughts
Relational database semantics are what runs the business. They are where the action is.
But those semantics must be improved, formalized, integrated and late-bound.
Specific projects/organizations must not be allowed to establish semantic silos. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 2 - 2
Part I : Part I © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 3 - 2 Basic Concepts
Controlled Vocabularies
The foundation for clear definitions.
Taxonomies
The backbones of ontologies.
Thesauruses
Translating searches for text strings into searches for concepts.
Ontologies
Data models, and business rules, on FOPL steriods.
Inference Engines
How ontologies are put to work.
Semantics
The meaning behind the data.
Semantic Interoperability
For federated queries: “Are we speaking the same language?”
Controlled Vocabulary: Definition : Controlled Vocabulary: Definition Controlled Vocabulary: a set of business terms precisely enough defined that it is clear what they do and do not apply to. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 4 -1
Controlled Vocabularies and Data Dictionaries : Controlled Vocabularies and Data Dictionaries Data Dictionary: a bad definition. Customer: a person or organization who has purchased or may purchase a product or service from our company. Data Dictionary: a good definition. Customer: a [person] or [organization] who, within the last five years, has either responded to {marketing material} or made an {invoiced purchase} from our company. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 5 -2
Taxonomy: Definition : Taxonomy: Definition Taxonomy: a hierarchical structure where the only semantic relationships are IS-A and INSTANCE-OF . Rows of tables are the INSTANCES OF those types. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 6 - 1
Using a Thesaurus: an Example : Using a Thesaurus: an Example That would include the text strings “SOA”, “Service-Oriented Architecture”, “Service Oriented Architecture”, “service-oriented architecture”, etc. A formal thesaurus would specify all these strings as synonyms. A search engine using that thesaurus, and given any one of those text strings, could search for the concept. A search for the string “SOA” in MS Word documents. But what we really want is to find all references to Service Oriented Architecture. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 7 - 2
Ontology: Definition : Ontology: Definition Ontology: “… an ontology is a data model that represents a set of concepts within a domain and the relationships between those concepts. It is used to reason about the objects within that domain” (Wikipedia). Uses of ontologies include: Ontological Commitment: what kinds of things we say there can be instances of. Thus, the tables in our databases. Their rows represent what we say exists. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 8 - 2
Thesaurus: Definition : Thesaurus: Definition Thesaurus:
A book of synonyms, often including related and contrasting words and antonyms.
(Webster’s on-line) Formal Thesaurus:
A thesaurus that supports automated inferencing, i.e. one that can be used by software. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 9 - 1
Ontologies and Relational Databases : In business IT, relational databases and SQL are state of the art in ontology management tools. They have strong extensional database management capabilities, but their support for intensional database features is minimal. Ontologies and Relational Databases Intensional Databases.
The core of each intensional database is a set of FOPL axioms which fully describe its subset of the enterprise data model.
All the information in the DBMS catalog, including entity, referential and domain integrity constraints.
Augmented with an FOPL-expressed terminological taxonomy of the entity, attribute and relationship names for the data model.
Including the state-transformational rules otherwise hardcoded in programs or stored procedures.
Also including potentially volatile ontological commitments otherwise hardcoded in database schemas. Extensional Databases.
What we know of as simply databases. The Claims Database, the Shop Floor Management Database, the Personnel Database, etc. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 10 - 3
Inference Engine: Definition : Inference Engine: Definition Inferencing: drawing conclusions from what you already know. More specifically, deriving a conclusion from a set of premises by means of deductive logic. Example:
John is older than Mary
Mary is older than Mike
John is older than Mike
(4) John loves Mary
(5) Mary loves Mike
(6) John loves Mike Inference Engine: software that, without hardcoding, will return (3) to a query, but not (6). Extensional/Fact Database: a database that contains (1), (2), (4) and (5). Intensional/Rules Database: a declarative set of rules that defines “older than” as a transitive (and anti-symmetric) relationship, and defines “loves” as an intransitive (and non-symmetric) relationship. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 11 - 3 Example:
[(1 & 2) 3]
BUT
~ [(4 & 5) 6]
Semantics: Definition : Semantics: Definition Semantics: in linguistics, what words and sentences mean. In IT, what data means. Example:
885.22. This is just data, a string of characters. We don’t know what it means.
$885.22. This is a string of characters representing a monetary amount denoted in U.S. dollars. But this meaning is too general to do us much good.
But in a Customer Invoice table, under the column total-amount, it means the amount due from that customer, for the products on that invoice’s line items. This $885.22 is still data. But it is data with meaning – some of that meaning formalized, but most of it not formalized. For example, to the DBMS, the label “total-amount” is just a character string associated with that column. The DBMS doesn’t “understand” what “total-amount” means. It can’t reason about it. But it can manipulate symbols (data) so as to produce the same results as we would produce by reasoning! This is what we mean when we talk about software “reasoning”, or “doing inferencing”. © Copyright 2007, Mindful Data, Inc. All rights reserved Slide 12 - 3
Semantics: Commentary : Semantics: Commentary In business IT systems – applications, databases, stored procedures, messages – semantics is spread out all over the place. Most of the transformation rules which govern inserts, updates and deletes are hardcoded, either in database triggers, or in application programs. Most of the structural rules which define entities, attributes and relationships are hardcoded in data schemas. If we consolidate these rules, express them declaratively, and late bind them, then: The result will be:
a codebase that is run-time bound to declarative rules which describe the constraints on transformations of business data.
a database that is run-time interpreted by reference to declarative rules which designate highly abstract schemas as specific business objects.
The combined set of these declarative rules is the intensional database. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 13 - 3
Semantic Interoperability: Definition : Semantic Interoperability: Definition Database Interoperability: two databases are interoperable if their data can be assembled by a single SQL statement. Syntactic Interoperability:
two atomic data objects are syntactically interoperable if SQL can JOIN on them.
two composite data objects are syntactically interoperable if SQL can UNION them. Semantic Interoperability:
two atomic data objects are semantically interoperable if the label and domain elements of one of them can be fully mapped, using formal semantic relationships, to the label and domain elements of the other.
two composite data objects are semantically interoperable if each composing object in one can be mapped, using formal semantic relationships, to a composing object in the other. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 14 - 3
Semantic Interoperability: Commentary : Semantic Interoperability: Commentary The Semantic Net is all about inter-company semantic interoperability. It’s about how software can “know” what data out there on the web really means.
Does my supply chain partner mean the same thing by “customer” that I do? (In other words, will a query ranging across both our Customer tables return a consistent result set, or a bag of apples and oranges?) But the greatest payback will be in improving intra-company semantic interoperability.
Why don’t sales totals ever add up exactly right at executive meetings?
Are individual loading docks locations, or just the buildings that contain them? (What do we mean by “location”?)
Is material manufactured by another plant in our company, which is also a leaf node on the bill of materials in our plant, really raw material? (What do we mean by “raw material”?) © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 15 - 2
Part II : Part II © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 16 - 1 Ontologies and Databases: an Action Plan.
Develop a rigorous semantics.
Clean definitions, paradigmatic and borderline examples.
Develop a formalized semantics.
Semantics that DBMSs and other inference engines can interpret.
Develop an integrated semantics.
Consistent vocabularies, taxonomies and ontologies across the enterprise.
Develop a late-bound semantics.
Semantics expressed declaratively, in a rules engine / intensional database.
Objective 1. Develop a Rigorous Semantics: Data Dictionaries and Controlled Vocabularies : Objective 1. Develop a Rigorous Semantics: Data Dictionaries and Controlled Vocabularies © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 17 - 1
Objective 1. Develop a Rigorous Semantics: Type Hierarchies and Taxonomies : Objective 1. Develop a Rigorous Semantics: Type Hierarchies and Taxonomies © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 18 - 2
Objective 2. Develop a Formalized Semantics: Database Schemas in FOPL : Objective 2. Develop a Formalized Semantics: Database Schemas in FOPL © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 19 - 3
Objective 3. Develop an Integrated Semantics : Objective 3. Develop an Integrated Semantics No semantic silos! The foundation is the semantics in the data that runs the company. Where is that data? In the production databases, both transactional and analytical. Once rigorously expressed, call the result the “enterprise semantic base”. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 20 - 2 Once formally expressed, call the result the “enterprise ontology”. After this is underway, and not before, is the time to look at industry ontologies, and even upper ontologies. Because now we will have our company’s “run the business” ontology, the foundation on which we will:
build up to industry ontologies, and then on to SUMO, Cyc or other upper ontologies; and
build out to other collections of data within our enterprise whose ontologies will likely extend our foundational ontology.
Objective 4: Develop a Late-Bound Semantics : Objective 4: Develop a Late-Bound Semantics Terminological taxonomies of the entities, attributes and relationships of the enterprise data model. Schemas from the database catalog. An intensional database, in which all “rules” are expressed declaratively, in FOPL, and are consequently late-bound to both code and physical schemas.
Which can therefore be used by a DBMS (or by software for managing unstructured data) to increase inferencing power.
Those rules include: Transformation rules extracted from hardcode. Synonyms and antonyms expressed as terminological equivalences or as term/negation pairs. Expressed declaratively, in the intensional/rules database, these semantics are late-bound to the DBMS and to any other inference engine. Fine-grained ontologies specified in database schemas, extracted from those schemas. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 21 - 3
Part III : Part III © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 22 - 7 Discussion How are taxonomies expressed in logical data models? How are ontologies expressed in logical data models? How do relational databases and the SQL language constitute an inferencing engine? What's different about the inferencing engines that are appearing as taxonomy and ontology management tools for business IT? How can my company get started benefitting from taxonomies and ontologies right away? How can my company get started preparing for the enhanced inferencing power of ontologically-informed DBMSs? What’s the first thing I should do when I get back to work next week?
Part IV : Part IV © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 23 - 1 Final Thoughts.
Relational database semantics are what runs the business. They are where the action is.
But those semantics must be improved, formalized, integrated and late-bound.
Specific projects/organizations must not be allowed to establish semantic silos.
Final Thoughts – 1 : Final Thoughts – 1 Relational database semantics are what runs the business. They are where the action is. These are the ontologies that will enable semantic interoperability with our suppliers, customers, regulators and other parties. © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 24 - 2 Do these two tables mean the same thing by product? By customer? By calendar? By invoice? By purchase order? By return? Etc, etc.
If we don’t know, we can’t coordinate with these other parties. Do they mean the same thing? Currently, the burden of insuring this falls primarily on people. If we can formalize the ontologies currently imperfectly expressed in our production databases, then the burden of insuring semantic interoperability will fall primarily on inference engines (like enhanced DBMSs).
Final Thoughts – 2 : Final Thoughts – 2 © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 25 - 1
Relational database semantics must be:
improved (Objective 1)
formalized (Objective 2)
integrated (Objective 3)
late-bound (Objective 4)
improve formalize Late-bind integrate
Final Thoughts – 3 : Final Thoughts – 3 © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 26 - 2 An enterprise semantics, based on a controlled vocabulary, taxonomy and ontology, is the first order of business.
Its source is the semantics already expressed in the enterprise’s run-the-business relational databases. Projects to apply semantic management to unstructured data – scanned and OCR’d documents, emails, pdf files, doc files, etc – must not be allowed to acquire and/or develop their own semantics.
They must not become ontology silos.
End of Presentation : End of Presentation © Copyright 2007, Mindful Data, Inc. All rights reserved. Slide 27 - 1 Thank you. If there’s anything else I can do to help, please give me a call. I hope this was helpful. Now it’s time to go back home and begin creating an enterprise semantics.