Open Conceptual Data Models

Making the Conceptual Layer Real
via
RDF Linked Data

By Kingsley Idehen ,
Carl Blakeley

Conceptual Data Models in the Linked Data Web

Linked Data Vision:

The transition of the Web
- from a Web of linked documents
- to a Web of interlinked structured data items (aka: entities, data objects, resources)

Concurrent trend in the IT industry:

The transition of the Web
- A recognition of the benefits of conceptual data models vs logical data models

The Big Question:

To what extent does the Linked Data support conceptual level data models?

Open Conceptual Data Models

Topics:

Conceptual & Logical Data Models
Conceptual Models for the Semantic Web
Realizing Conceptual Models through Ontologies & Linked Data
Virtuoso RDF Views
ADO.NET Data Services & the Entity Data Model

Conceptual & Logical Data Models

Describe a software system's target problem space

Typically, in today's database-driven applications

Three levels of data model

Physical
- How data is physically represented on disk
Logical (aka logical schema)
- Expresses problem domain in terms of data management technology (tables / columns)
- e.g. relational schema
Conceptual (aka conceptual schema)
- Purely semantic description of problem space
- Describes things (entities), their characteristics (attributes) & associations between things (relationships)

Logical Data Model

Most prominent of the three data model types
Main focus of database applications
- Due to pervasiveness of SQL in application code

Weaknesses

Impedance mismatch
Loss of semantics during development process
Heterogeneous databases & interoperability

Logical Data Model Weaknesses

Impedance Mismatch

SQL expresses queries in terms of tables / views
- => targets logical schema
Normalization fragments the data model
- Entities & their attributes may be split across several tables
- Navigation between objects requires relational joins over two or more tables
- Table rows must be reconstituted into higher level conceptual entities

Conceptual level data model is desirable to:

Remove impedance mismatch
Isolate application from changes to logical data model
Provide framework for human level interaction

Logical Data Model Weaknesses

Loss of Semantics During Development

Process:

Develop conceptual model (E-R modelling)
Transform to logical model for implementation
Derive physical model from logical model

Problems:

Each move to a lower level model discards meaning
Higher level model typically not retained
Model semantics fragmented across schema / business rules / application code
Application must know logical data model
- Must be hardcoded or inferred (imperfectly) from system tables

Logical Data Model Weaknesses

Heterogeneous Databases & Interoperability

Logical data model

Describes problem domain in terms of tables/columns
Requires SQL to navigate model

Application

Exposed to specifics of a particular vendor's RDBMS In heterogeneous database environment, must handle
Different SQL dialects
Different schemas
- No explicit data model.
- No explicit semantics.
Interoperability/integration = perpetual problem for IT depts

Conceptual Models for the Semantic Web

Growing recognition in the industry of the benefits of a conceptual, rather than logical, model for data-centric applications
- e.g. Microsoft's Entity Data Model / Entity Framework
Semantic Web technologies provide powerful tools for this paradigm shift

Benefits of Conceptual Models

How the Semantic Web benefits

More faithfully represents human view of domain of interest
Conceptual model & semantics
- Explicit & available globally
- Not implicit & fragmented across business logic / UI etc
Better / explicit semantics promises better search engines
Much easier heterogeneous data integration
- Data on the Web is inherently heterogeneous

Application Areas - Present & Future

Social networking, e-commerce, collaborative working
- Require shareable, standards-based, cross-platform conceptual views of data
Data portability
- Needed as Web users maintain multiple points of presence - blogs, social network accounts etc.
Open business models
- Require exchange & integration of large amounts of data
Scientific research - sharing of knowledge & findings
- Requires transparent access to distributed heterogeneous data
- Requires database integration using global schema
Autonomous intelligent agents
- Free humans from large-volume information processing

Semantic Web Technology Benefits

What Semantic Web technologies bring:

Ontologies

Can represent common semantics
- Spanning databases, applications, enterprises, on-line communities
Act as a shared conceptual model
Provide common models (FOAF, SIOC etc)

Common Semantics (Ontologies) & Common Data Representation (RDF)

Enable cross data source querying using SPARQL
- Content from several sites can be combined / explored
- Querying using proprietary APIs unnecessary
- Brute force data merging unnecessary

Open Data Formats, Platform Independence, Common Models

Allow data portability and data integration

Realizing Conceptual Models

Ontologies

Provide the building blocks of Semantic Web conceptual models
Define the concepts and their relationships in a domain of interest

Describing Classes & Properties - Ontology Languages

RDFS
- Introduces the notions of concepts (classes) & instances
OWL
- Adds more vocabulary for describing:
  - relations between classes
  - cardinality
  - richer typing of properties, etc.

Goodness of Fit

RDF was designed from the ground up as a metadata data model
RDF / RDFS / OWL work directly at the level of conceptual models
Conceptual model terminology matches RDF/OWL terminology
- Concepts, entities, attributes, relationships

A natural fit!

RDF lends itself naturally to describing conceptual models

Semantic Expressivity

DDL-based Relational Model

Relationship between two entities isn't explicit
Foreign key relating two rows in separate tables doesn't express the nature of the relationship
Semantics must often be inferred from table definitions

RDF-based Conceptual Model

Relationship between two entities is stated explicitly by predicate in subject-predicate-object triple
Semantic expressivity of RDF/RDFS/OWL is much better than DDL
Has richer semantic content than equivalent DDL-based logical/relational model

RDF Conceptual Model - Artist / Records / Tracks

RDF Conceptual Model

Global Granular Information Sharing

Traditional Logical/Relational Data Model

Schema described by DDL is internal to DBMS
Primary keys identifying an individual table row (i.e. entity instance) not globally unique, not easily usable outside host DBMS
Gives rise to 'data silos'

RDF's use of HTTP-based URLs

Externalises the data and schema
Makes both globally accessible & scalable
Provides globally unique IDs for entities/relations/classes
A vehicle for granular, global information sharing down to the equivalent of the record level

Linked Data - What is It?

A method for exposing, sharing & connecting data on the Web

A term coined by Tim Berners-Lee that describes HTTP-based Data Access by Reference for the Web
Open Data Access & Connectivity mechanism for the Web
A richer linking mechanism for the Web that takes us from Hypertext Links (Document to Document) to Hyperdata Links (across things that documents are about)

Linked Data - Why Is It Important

It exposes the compound nature of Web Resources
- Information resources (Containers) are uniquely identified & referenceable
- Entities within Containers are uniquely identified & referencable
It provides an Open Data Access & Connectivity mechanism for the Web
It delivers a powerful mechanism for meshing disparate and heterogeneous data sources

Linked Data Model

Changes the focus from linked documents to linked entities
The document as a data container becomes less relevant

Linked Data Model

Hyperdata Links Between Data Objects

Linked Data Benefits - Natural Navigation

Natural Navigation Through Typed Links

RDF entities are identified by dereferencable URIs (URLs)
Navigating from one data item to another is easy
- One click to dereference in Semantic Web Browser
- e.g. OpenLink Data Explorer
URI of object in an RDF statement is a typed link
- Link's "type" is defined by the statement predicate

Relational/Logical Model

Cumbersome
Requires SQL joins + typically Object-Relational mapping
e.g. in C#: track = lennonAlbum.Tracks["Imagine"]

Linked Data Benefits - Aggregatable Data

Often desirable to have an integrated view of all the data available about an item or topic

Database Realm

Integration problematic, difficult to combine logical schemas

Semantic Web

Data aggregation is easy: every resource has a unique URI
- Individual items can be linked
- Conceptual models can be linked
Cross-domain links enrich domain knowledge
Different facets of the same entity may be described by different URIs minted by different authors
- Can be linked. e.g. owl:sameAs, rdf:type predicates
- May expose facts not directly represented in any one source

Linked Data - Data Aggregation

Linked Data Benefits - Self Describing Data

RDF

A technology for creating self-describing Web resources
Entity's type definition 'accompanies' it using rdfs:type
An RDF dataset can be queried using SPARQL without knowing anything beforehand about the data
Provides the basis for powerful data exploration tools

Logical / Relational Schema

Users / applications need a detailed understanding of the schema to use and navigate the data
Application's knowledge of the schema typically hardcoded
Ad-hoc end-user data exploration potentially error prone

Linked Data Benefits - SPARQL

If a user agent has no built-in knowledge of a particular RDF subject, predicate or object, it can use the URI to retrieve the information

The Power of SPARQL

Discover what sorts of things a data source contains

select distinct ?URI ?ObjectType where { ?URI a ?ObjectType }

Determine all the properties of an entity class

select * where { <http://my.org/resourceTypes/Department> ?property ?hasValue }

Determine all the properties and values of an entity instance

DESCRIBE <http://my.org/resource/Accounts>

No prior knowledge of the RDF data source is needed

Virtuoso - Linked Data Generation Options

Conceptual layer insulates Linked Data consumers from RDFization infrastructure & data source heterogeneity

Virtuoso - Linked Data Generation Options

Virtuoso RDF Views

Expose relational data as RDF
Provide the means to move from a logical model view to a conceptual model view

Available for querying through SPARQL or SPASQL (SPARQL embedded in SQL)
No physical regeneration of relational data

RDF Views =
- Virtuoso RDF Meta-Schema +
- Meta-Schema Language
MSL =
- A domain specific, declarative language for mapping a logical SQL data model to a conceptual RDF data model

Northwind Demo Database: RDF View Definition Extract

Demo.demo.Customers:

Northwind RDF View Definition

prefix northwind: <http://www.openlinksw.com/schemas/northwind#>
...
create iri class northwind:Customer
<http://^{URIQADefaultHost}^/Northwind/Customer/%U#this>
(in customer_id varchar not null)
...
alter quad storage virtrdf:DefaultQuadStorage
...
from Demo.demo.Customers as customers
from Demo.demo.Orders as orders ...{

create virtrdf:NorthwindDemo
as graph iri ("http://^{URIQADefaultHost}^/Northwind") {
...
northwind:Customer(customers.CustomerID) a foaf:Organization
as virtrdf:Customer-CustomerID ;
northwind:companyName customers.CompanyName as ... ;
...
northwind:fax customers.Fax as virtrdf:Customer-fax .

northwind:Customer(orders.CustomerID)
northwind:has_order northwind:Order(orders.OrderID)
as virtrdf:Order-has_order .
... }}

Northwind Demo Database: Customer Table to RDF Entity Mapping

Customer Table to RDF Entity Mapping

Customer Table to RDF Entity Mapping Diagram

LinqToRdf + Virtuoso

LinqToRdf to MusicBrainz - Conceptual Model Veneer

ADO.NET Data Services & Entity Data Model

A framework for exposing 'pure data' service over HTTP

No support for RDF

Fails to imbibe any of RDF's inherent benefits

Lack of platform independence & standards compliance

Supports REST-style interfaces
Supports Atom, JSON and XML payloads

But

Server-side: Windows only
Consuming Astoria services at a higher level requires Windows .NET client or Silverlight-supported browser

ADO.NET Data Services & Entity Data Model

Server-side only conceptual model

Powerful URL addressing to query/navigate/sort/filter etc
- Customers collection: http://myserver/data.svc/Customers
- Customer ALFKI: http://myserver/data.svc/Customers('ALFKI')
- Customer ALFKI's orders: http://myserver/data.svc/Customers('ALFKI')/Orders

But

Client must know conceptual schema
- e.g. to construct above URIs

Lack of Deferencable Entity IDs

Ability to discover entities and dereference their descriptions (attributes/relations) is confined to the facilities offered by .NET

c.f. SPARQL's ability to handle unknown data sources

ADO.NET Data Services & Entity Data Model

No Support for Non-SQL Data Sources

Astoria is aimed exclusively at making relational data Web accessible

c.f. Semantic Web & Linked Data

Recognize that vast amounts of data resides in unstructured and semi-structured data sources
Support for embedding RDF into existing (X)HTML
- RDFa, GRDDL, eRDF
Emerging tools for converting non-RDF data to RDF
Emerging tools for exposing SQL data as RDF

Astoria lacks scalability & scope of Semantic Web technologies