ontop: A tutorial

+
ontop: A Tutorial
Mariano Rodriguez Muro, Ph.D.
Free University of Bozen-Bolzano
Bolzano, Italy
http://www.rodriguez-muro.com
Protégé-OWL Short Course
September 2-4, 2013
Vienna, Austria

+
Disclaimer
License
This work is licensed under the
Creative Commons Attribution-Share Alike 3.0 License
http://creativecommons.org/licenses/by-sa/3.0/
The material for this presentation is available at:
https://www.dropbox.com/sh/q3aowgiq5dnco7n/as0QniGPKy

+
Who am I?
 Researcher at:
Free University of Bozen Bolzano,
Bolzano, Italy
From October at:
IBM Watson Research Center
 Research topics: OBDA, Efficient reasoning in OWL, Query
rewriting, Data integration
 Leader of the ontop project

+
Why are we here?
 Data and ontologies
 To get the basics of ontology based data access
 To learn how to do it with ontop
 To grasp some of the possible uses of the technology and
hint to the resources available

+
Tutorial Overview
 Part 1: Introduction
 Quick Introduction to Ontology Based Data Access
 Part 2: The basics
 Creating an SQL database, creating simple (direct) mappings with
ontopPro and querying.
 Part 3: Modeling in OBDA
 Creating mappings that reflect the domain
 Part 4: Data Integration
 Using ontop to query data from multiple sources
 Part 5: Ontologies and ontop
 Extending and using with domain knowledge (OWL)

+
Material
 The tutorial is organized as a hands-on session. Try to perform
the described tasks.
 Most command/mappings/queries are in files in the “materials”
folder, still, try to write them on your own
 Material included (ontop-tutorial-viena13.zip)
 README.txt is an index for the ZIP file
 H2 Database (h2.zip)
 .obda/.owl files (resulting mappings and ontologies for all examples)
 .sql files (. SQL commands that create the tutorial DBs)

+
SQL DBs
 Standard way to store LARGE volumes of data
 Mature, Robust and FAST
 Domain is structured as tables, data becomes rows in these
tables.
 Powerful query language (SQL) to retrieve this data.
 Major companies developed SQL DBs for the last 30 years
(IBM, Microsoft, Oracle) and even open source projects are
now quite robust (MySQL, PostgreSQL).

+
OBDA and motivation
 Ontology Based Data Access (OBDA) is an research are that
focuses on accessing data through ontologies. –ontop-’s focus is
on SQL DBs (RDBMs)
 Benefits:
 Flexible data model (OWL/RDF)
 Flexible query language (OWL or SPARQL)
 Inference
 Speed, volume and features (by reusing SQL DBs)
 Possible applications
 Semantic Query Answering
 Data integration
 Semantic Search

+
Two approaches for OBDA
Extract Transform Load (ETL)
Reasoner
Source
Application
TBox
Inputs
Data Code
Data is transformed into OWL ABox
assertions that are combined with
OWL axioms and then given to a
reasoner or query engine.
Limitations: performance and memory.

+
Two approaches for OBDA
On-the-fly
Reasoner
Source
Application
Ontology
Mappings
Input
Mappings are axioms that relate the data in a
RDBMs to the vocabulary of the ontology (the
classes and the properties), they “connect” the two
vocabularies in a sense.
The input are ontology and mappings,
the reasoner answers the queries by
transforming them into queries over
the source.
The reasoner is connected to the
source, data is not duplicated, is always
up-to-date.

+
 ontop is a platform to query RDBMs through OWL/RDFS
ontologies on-the-fly, using SPARQL. It's extremely fast and is
packed with features
 It‟s composed by 2 main components:
 Quest. A reasoner/query engine that is able to answer SPARQL 1.0
queries, supports OWL 2 QL inference and a powerful mapping
language. Can be run in Java applications or a stand-alone
SPARQL server.
 ontopPro. A plugin for Protégé 4 that provides a mapping editor,
and that allows to use Quest directly from Protégé.
 Today we will focus learning to use OBDA with ontopPro

+
Part 2
ontopPro: The basics
SQL, Mappings, Queries

+
Overview
 Flash recap of SQL DBs with
H2
 Using ontopPro
 Connecting a DB with
Protégé
 Creating mappings
 Querying
 About mappings in ontop
 About query answering in ontop

+
An SQL database, H2
 A pure java SQL database
 Easy to install
 Just unzip the downloaded package, already in your USB stick h2-
simple.zip
 Easy to run, just run the scripts:
 Open a terminal (in mac Terminal.app, in windows run cmd.exe)
 Move to the H2 folder (e.g., cd h2)
 Start H2 using the h2 scripts
 sh h2.sh (in mac/linux)
 h2w.bat (in windows)

starting H2 with the
terminal
A yellow icon indicates the DB is
running (right click for a context
menu)

jdbc:h2:tcp://localhost/test
jdbc:h2:tcp: = protocol information
locahost = server location
test = database name

+
Creating the database
 We‟ll create a table to store lung cancer information as follows:
patientid name type stage
1 Mary false 2
2 John true 7
type is:
• true for Non-Small Cell
Lung Cancer (NSCLC)
• false for Small Cell Lung
Cancer (SCLC)
stage is:
• 1-6 for stage I,II,III,IIIa,IIIb,IV
NSCLC
• 7-8 for Limited,Extensive SCLC

+
Creating the table
 To create the table (file patient-table1.sql):
CREATE TABLE tbl_patient (
patientid INT NOT NULL
PRIMARY KEY,
name VARCHAR(40),
type BOOLEAN,
stage TINYINT
)

+
Inserting the data
 To insert the data:
INSERT INTO tbl_patient (patientid,name,type,stage)
VALUES
(1,'Mary',false,2),
(2,'John',true,7);

+
Retrieving data: SQL
 To retrieve all data: SELECT * FROM TBL_PATIENT ;

+
Other relevant queries
 To get all id‟s of patients with NSCLC
SELECT patientid FROM TBL_PATIENT WHERE TYPE = false
 To get all information about patients with NSCLC and stage 3
or above
SELECT patientid FROM TBL_PATIENT
WHERE TYPE = false AND stage >= 2

+
First ontop mapping
 Objective
 That each row generates the following OWL data
 An OWL individual of the form:
:db1/1
 OWL assertions of the form:
ClassAssertion( :Person :db1/1 )
DataPropertyAssertion ( :id :db1/1 “1”)
DataPropertyAssertion ( :name :db1/1 “Mary”)
DataPropertyAssertion ( :type :db1/1 “false”)
DataPropertyAssertion ( :stage :db1/1 “2”)
 That is, we define a vocabulary of Classes and Properties that
we want to “populate” using the data from the database.
A direct mapping

+
A direct mapping (cont.)
 Seen graphically:
 Things to note:
 The OWL object is identified by
an IRI
 Values have OWL data types
patientid name type stage
1 Mary false 2

+
Step 0: Starting Protégé+ontop
 Unzip the protégé-ontop bundle from your material
This is a Protégé 4.3 package that includes the ontop plugin
 Run Protégé using the run.bat or run.sh scripts. That is,
execute:
cd Protege_4.3_ontopPro/
sh run.sh

+
Step 1: Defining the base URI
 Define the ontology base URI: http://example.org/
 Save the ontology
 Close and re-open Protégé
(Sorry this is due to a bug)
 Enable the OBDA Model tab in
Window -> Tabs

+
Step 2: Add the datasource
 Using the OBDA model tab, we now need to define the
connection parameters to our lung cancer database
 Steps:
 0. Switch to the OBDA model tab
 1. Add a new data source (give it a name, e.g., LungCancerDB)
 2. Define the connection parameters as follows:
 Connection URL: jdbc:h2:tcp://localhost/test
 Username: sa
 Password: (leave empty)
 Driver class: org.h2.Driver (choose it from the drop down menu)
 3. Test the connection using the “Test Connection” button

+
 A “Connection is OK” means Protégé and ontop were able to
connect to our H2 server and see the “tests” DB we just created.
We are now ready to add the mappings for the DB.

+
Step 3: Create a mapping
 Add the class:
http://example.org/Patient
 Switch to the “Mapping Manager” tab in the OBDA Model tab.
 1. Select the LungCancerDB source
 2. Add a mapping with ID “patient-map”
target: :db1/{PATIENTID} a :Patient .
source: SELECT * FROM TBL_PATIENT
NOTE: use upper case

+
Adding a Mapping
 Select the LungCancerDB from the drop down menu.
 Click the “Create Button”

+
 The “Assertion template” a.k.a. “triple Template” tells ontop how
to create URI‟s and Class and Property assertions using the
data from the DB (from the SQL query)

+
The meaning of mappings
 Mappings + DB data “entail
(consequence)” OWL data,
i.e., OWL ABox assertions.
 These “entailed” data is
accessible during query time
(the on-the-fly approach) or
can be imported into the OWL
ontology (the ETL approach)

+
The meaning of mappings
 ontop‟s main way to access data is on-the-fly, however, you can
also do ETL using the “import data from mappings” function
in the “ontop” menu.
 Do it now and explore the result in the “individuals tab”, when
done remember to delete these individuals.
Use with care, you may run out of memory.

+
On-the-fly access to the DB
 This is the main way to access data in ontop and its done by
querying ontop with SPARQL.
 The “query engine”/”reasoner” that comes with ontop is called
“Quest”
 Enable Quest in the “Reasoner” menu

+
On-the-fly access to the DB
 Next, enable the “OBDA query” tab (ontop SPARQL) in the tabs
menu

+
Querying with Quest
In the OBDA Tab:
1. Write the SPARQL
query
2. Click execute
3. Inspect the results

TEMPLATE: :db1/{PATIENTID} a :Patient .
The result is no
longer numeric
ID‟s in the
database. The
results are URI‟s
constructed in
the way that you
wrote in the
mapping by
replacing the
“column
references” with
the actual values
obtained from the
database (the
values in each
row)

+
The rest of the patient mappings
 Add the following Data
Properties:
 :id
 :name
 :type
 :stage
target: :db1/{PATIENTID} :id {PATIENTID} .
target: :db1/{PATIENTID} :name {NAME} .
target: :db1/{PATIENTID} :type {TYPE} .
To complete the model we can add the following
mappings one by one and “synchronize” the reasoner:
target: :db1/{PATIENTID} :stage {STAGE} .
OR…

+
The rest of the patient mappings
 Add the following Data
Properties:
 :id
 :name
 :type
 :stage
target: :db1/{PATIENTID} a :Patient ;
:id {PATIENTID} ;
:name {NAME} ;
:type {TYPE} ;
:stage {STAGE} .
Or, you can modify the original mapping as follows so that
it generates multiple assertions at the same time:
Don‟t forget to synchronize with the reasoner…

+
About Mappings
 A mapping represents OWL assertions, one set of OWL
assertions for each result row returned by the SQL query in the
mapping. The assertions that the mapping represents are those
obtained by replacing the place holders with the values from
the DB.
 Mappings are composed by:
 Mapping IDs are arbitrary names for each mapping (choose
something that allows you to identify the mapping)
 The “Source” of the mapping is an SQL query that retrieves some
of the data from the database.
 The “Target” of the mapping is a form of “template” that indicates
how to generate OWL Assertions (class or property) in a syntax very
close to “Turtle” syntax for RDF.

+
Assertion Template Examples
 Assertion templates are formed as a triple
“subject predicate object”
 The subject is always a URI, the object maybe another URI or an OWL value.
 Class Assertions use rdf:type or a as predicate, and a URI as object (the
class name) e.g.,
:db1/{id} rdf:type :Person
<http://live.dbpedia.org/page/{name}> a :Writer
 Object/Data Property Assertion have any URI as predicate (the property
URI) and a URIs or OWL Value as object
:db1/{id} :name {NAME}
:db1/{id} :age {C1}^^xsd:string
:db1/{id} :knows :db1/{id2}
:db1/{id} :knows :Michael_Jackson

+
Practical Notes About Mappings
 With ontopPro, mapping and data source definitions are
stored in .obda files
 .obda files are located in the same folder as the .owl ontology
 They should be named as the .owl file
 .obda files are text files, they may be edited and created
manually, this can be more convenient in several cases, e.g.,
automatically generating large amounts of mappings, quick
refactoring using regular expressions, etc.

+
About Query Answering in Ontop
 ontop‟s query engine uses “query rewriting” techniques
 Given a SPARQL query, ontop translates it into an SQL query
using the mappings (and the ontology). You can get the SQL
query generate by ontop using the context menu in the OBDA
query tab.

+
About Query Answering in Ontop
 Key features:
 Volume: By relying on SQL DBs, the datasets that ontop can handle
are in the GBs and TBs
 Fast: ontop generates efficient SQL queries, that when combined
with a fast SQL engine to provide answers in ms. Not all SQL
queries are fast, most of the research and development efforts in
ontop go towards generating FAST SQL queries
 Possible drawbacks:
 Maturity: SPARQL support in ontop is under development and
many features are still missing
 Know-how: SQL expertise will be required to obtain the best
performance with large datasets

+
Part 3
Domain Modeling in OBDA

+
 The OWL vocabulary so far is a one-to-one reflection of the
database, not very interesting or useful
 We would like:
 Application independent
vocabulary
 Vocabulary beyond the
one explicit in the DB
 Individuals and relations between them
that reflect our understanding of the
domain
 For example…
Application independent mappings

+
Redesigning our model
Highlights:
• The vocabulary is more domain oriented
• No more values to encode types or stages. There is
a new individual :db1/neoplasm/1 that stands for the
cancer (tumor) of Mary and it is an instance of the
class :NSCLC. There are URI‟s (individuals) that
represent the stage of the cancer
This model is closer to the formal model of the domain, independent from
the DB. Later, this will allow us to easily integrate new data or domain
information (e.g., an ontology).

+
Constructing the new model 1
Remove the old mappings and vocabulary, then:
 Create the new vocabulary
Object Properties:
:hasStage, :hasNeoplasm
Classes:
:SCLC, :NSCLC
 Add mappings for the new classes and properties as follows:

+
Basic mappings
Basic mapping to generate the patient individual as well
as the new “neoplasm” for that individual.

+
Classifying the neoplasm
Now we classify the neoplasm individual using our knowledge of the
database.
We know that “false” in the table patient indicates a “Non Small Cell Lung
Cancer”, so we classify the neoplasm as a :NSCLC. Similar for :SCLC

+
Associating a stage
We associate the
neoplasms of each patient
to a stage. Note that the
stage is no longer an
arbitrary value, but a
constant URI with clear
meaning, independent from
the DB.

+
Querying the new model
In the new model now we can obtain the information of each patient and their
condition through URIs of classes or individuals that have clear semantics, not
DB dependent. We are using a “global vocabulary”.
This will allow us to easily integrate new DBs and a domain ontology…

+
Part 4
Data integration
Integration by alignment

+
Data integration in OBDA
 Even if two databases contain data about the same domain,
integrating the data is often problematic since the data may be
represented in different ways
 However, if proper modeling is used, integrating multiple data
sources using OBDA may become simple:
 Insert the data in the database (either as new tables, or through
database federation)
 Create the new mappings for the source such that they match the
“global vocabulary”
 Query using the global vocabulary as usual
 Consider for example…

+
A different Lung Cancer DB
 Consider a new lung cancer DB as follows (create it NOW in
H2 using the commands in patient-table2.sql)
ID name ssn age
1 Mariano SSN1 33
2 Morgan SSN2 45
ID stage
1 i
ID stage
2 limited
T_NAME
T_NSCLC
T_SCLC
In this DB information is distributed in
multiple tables. Moreover, the way in
which meaning is encoded is different.
In particular,
• The type of cancer is separated by
table
• The stage of cancer is text
(i,ii,iii,iiia,iiib,iv, limited, extensive)
Moreover, the IDs of the two DBs
overlap (ID 1 is a different patient
here, not Mary) and ssn and age do
not exist in the DB1

+
Basic mappings 2
 The URI‟s for the new individuals differentiate the data sources
(db2 vs. db1)
 Being an instance of NSCLC and SCLC depends now on the
table, not a column value
You can find this mappings in lung-
cancer3-4tables.obda

+
Stage mappings 2
The new mappings
reflect what we know
of this new data source.

+
The integration result
Now, using a single SPARQL query, we can query both data sources
independently of their structure; they have been aligned to a global view.

+
However
 Multiple sources maybe have different properties
 We cannot know before hand if we don‟t know the sources and
the only thing you see is the ontology
 This can be a BIG issue for the user of our integrating ontology,
since many queries would be empty, e.g.:
SELECT ?x ?y ?z WHERE {
?x a :Person ; :name “Mary” ; :ssn ?y ; :age ?z .
}
This query is empty because Mary is from DB1 and
individuals from DB1 have no SSN or AGE. Similar
problems arise with SQL DBs.
But, in SPARQL we have DESCRIBE…

+
Flexible queries with SPARQL
“Retrieve all information about
individuals named „Mary‟ and all
information about all conditions they
have”
PREFIX : <http://example.org/>
DESCRIBE ?x WHERE {
{?x :name ”Mary" .}
UNION
{ ?y :name ”Mary";
:hasNeoplasm ?x }
}

+
Flexible queries with SPARQL
“Retrieve all information about
individuals named „Mary‟ and all
information about all conditions they
have”
DESCRIBE ?x WHERE {
{?x :name ”Mariano" .}
UNION
{ ?y :name ”Mariano";
:hasNeoplasm ?x }
}

+
OBDA for Data Integration
 Key features of on-the-fly OBDI (ontology-based data
integration) with ontop:
 Flexible: Mapping and ontology languages are powerful enough to
accommodate most needs (consider that SQL allows even to
transform the data, make calculations, etc.)
 Dynamic: Changes in the data are automatically reflected during
query answering. Through DB federation new databases can be
incorporated easily
 Possible drawbacks:
 Performance: With large volumes of data (hundreds of thousands
of rows) performance may suffer (depends on the DB engine,
indexes, and other SQL related issues)

+
Data integration resources
You may integrate any JDBC resource, here go some interesting
options:
 Teiid – can integrate different DB SQL dbs and other types of
documents (XML, Excel, etc.)
http://www.jboss.org/teiid/
 Oracle database links – integrates Oracle DBs
http://docs.oracle.com/cd/B28359_01/server.111/b28310/ds_conce
pts002.htm
 MySQL Federated tables – integrates MySQL dbs
http://dev.mysql.com/doc/refman/5.0/en/federated-storage-
engine.html
 Excel as SQL – Integrates Excel spread sheets
http://sourceforge.net/projects/xlsql/

+
Domain knowledge
 Up to know, we only have “explicit data”, however, combining
data with domain knowledge (ontology) we can enrich our
queries with “implicit data”.
For example, that NSCLC is a kind of malignant tumor
(neoplasm), that having a neoplasm is a kind of condition, etc.
 This knowledge can be expressed using OWL axioms, which
ontop will use during query answering.

+
Lung Cancer knowledge
Terminology Knowledge (TBox)
Plus:
ObjectPropertyRange( :hasStage :LungCancerStage )

+
The result
 After synchronization, all the
implied information is available
during query answering
DESCRIBE ?x WHERE {
{?x :name "Mariano"}
UNION
{ ?y :name "Mariano" ;
:hasNeoplasm ?x }
}

+
Our data before (only explicit)
Our existing data looked like this picture.
With the new axioms, it now looks like this
(next slide):

+
Our data now (implicit and explicit)
Recall, in the on-the-fly approach
all this information is
available at query time but not
really stored anywhere

+
Domain Knowledge
 Large amounts of data “belong” in databases, i.e., it changes
fast, application specific, large volumes, etc.). OBDA allows you
to do keep it in the DB, but…
 Some data in the domain does belong on the ontology side,
i.e., static information, independent from the application. For
example:
ABox data

+
Domain Knowledge
 This information usually given in the form of OWL individual
assertions (ABox).
 A unique feature of ontop is its ability to mix these two worlds,
Allows to link virtual individuals to real individuals to achieve
things like:
ABox data
We want to do this for all
individuals in db1 and db2!

+
Hybrid ABoxes: How?
 Add the individuals ABox assertions to your ontology (6
individuals and 6 ABox assertions)

+
Hybrid ABoxes: How?
 Add the individuals ABox assertions to your ontology (6
individuals and 6 ABox assertions)
 Add mappings that link your “virtual” individuals to the real ones

+
Hybrid ABoxes
DESCRIBE ?x WHERE {
?x :age ?age ;
:recordIn [ a :ResearchCenter; :locatedIn [ :partOf :usa ] ]
FILTER (?age > 40)
}

+
Notes about reasoning in ontop
 ontop can only understand OWL 2 QL axioms, that‟s:
 subClassOf, subPropertyOf, equivalence
 InverseOf
 Domain and Range
 Plus some limited forms of qualified existential restrictions
 Any axiom that is not understood by ontop is ignored while
reasoning
 Reasoning is also done by means of query rewriting (no data
moves from the database). Again, most of our research goes
into generating efficient SQL.

+
Conclusions
Pointers and Final thoughts

+
Other features of ontop
 Mapping Assistant (OBDA model tab) - A view to help you
generate custom mappings quickly
 Mapping bootstrapping (OBDA menu) – automatically
generate “direct” mappings (actually the first mappings we
created can be generated automatically with this function)
 Mapping materialization (OBDA menu) – generate OWL
assertions from mappings with one click (import). Try it now, all
the data will be available in the “Individuals tab” and you can
now use it with any reasoner
 SPARQL end-point – Use the mappings and ontology
independently from Protégé, as a SPARQL server

+
Other features of ontop
 OWLAPI and Sesame– Once you created the ontology and
mappings, program your application with ontop and Java using
these.
 Command line tools – All previous features can be used directly
from the command line with ontop scripts
 R2RML mappings – Ontop now supports also R2RML mappings
(http://www.w3.org/TR/r2rml/) the expressive power of these more
or less the same, however our syntax is more user friendly ☺, use
R2RML for mapping exchange
 JDBC sources – Ontop can support any JDBC data source. This
means not only RDBMs, but anything that can be seen as a
RDBMs and queried with SQL, currently there are many wrappers
that allow to do this for Excel files, XML documents, etc.

+
Disclaimer
Although the code of ontop is evolving fast, there are several
(Sept/13) important issues to consider when using ontop:
 Datatypes many data types not supported yet, issues with
dateTime
 SPARQL Current target SPARQL 1.0 plus most features of 1.1 (no
paths). From SPARQL 1.0 we still miss several built in functions
 SQL issues Some issues with SQL and some DBs, e.g., problems
getting DB metadata, issues with caps and quotes to qualify
column names
 SQL Optimization Performance good , but could be better. Many
planned optimizations not yet implemented.
 GUI/ontopPro Many bugs in the GUI (we focus on the DB aspect)

+
Additional material
 Ontop‟s website
 http://ontop.inf.unibz.it
 Ontop‟s documentation
 https://babbage.inf.unibz.it/trac/obdapublic/wiki
 Ontop‟s source code
 https://github.com/ontop/ontop/
Since August‟13 ontop is a open source (AGPL).
Consider contributing!
 Ontop‟s google group
https://groups.google.com/d/forum/ontop4obda

ontop: A tutorial

More Related Content

What's hot

Viewers also liked

Similar to ontop: A tutorial

More from Mariano Rodriguez-Muro

Recently uploaded

ontop: A tutorial

Editor's Notes