KAON2: Scalable ABox Reasoning

1
KAON2Scalable ABox Reasoning

• Boris Motik
• PhD thesis advisors
• Rudi Studer (University of Karlsruhe)
• Ulrike Sattler (University of Manchester)

2
Problems with ABox Query Answering

• ABox query answering
• very important in the Semantic web
• inefficient in existing tools
• What are the main problems?
• to compute answers to a query Q, reasoners
  usually need to
• consider each possible answer (a1, , an)
• check whether assuming ?Q(a1, , an) leads to a
  contradiction
• this is clearly inefficient
• We should to compute the answers constructively

3
What Works in Practice?

• (Deductive) databases!
• optimized to handle large data quantities
• efficient constructive query answering algorithms
• Numerous optimizations were developed
• join-order optimization
• magic sets
• The main goal of KAON2
• Realize query answering in OWL byreusing
  (deductive) database!

4
New reasoning Algorithm

• This is important because
• one can equivalently evaluate queries in DD(KB)
• while doing so, I can use existing efficient
  techniques
• indexing
• magic sets
• identifies the part of the database relevant to
  the query
• join-order optimizations
• identifies the most efficient access path to data

OWL Knowledge Base KB
Disjunctive DatalogProgram DD(KB)

• Important property
• DD(KB) and KB entailthe same ground facts

5
The Architecture of KAON2
KAON2
Ontology API
Reasoning API
DIG API
Files
RDBMS
Reasoning Engine
ABox assertions
TBox axioms
Reduction into Disjunctive Datalog
Ontology Clausification
Disjunctive Datalog Engine
clauses
rules
http//kaon2.semanticweb.org/
6
KAON2 Demo
7
Performance Evaluation

• Query answering KAON2 is one or more orders of
  magnitude faster than existing tools
• RACER and Pellet
• TBox reasoning KAON2 is slower
• it can still solve nontrivial problems
• possible room for improvement

8
HermiTScalable TBox Reasoning

• Boris Motik
• Rob Shearer
• Ian Horrocks

9
Problems with TBox Reasoning

• Biomedicine typical application domain
• Biomedical ontologies are quite complex
• consist of many concepts
• concepts are highly interconnected
• Such ontologies cause problems
• cannot be handled using existing reasoners

10
Problem 1 Large Models

• Ontology
• A heart contains the left ventricle.
• The left ventricle is adjacent to a septum.
• The septum is a part of the heart.

• During reasoning, a reasoning creates a model of
  the ontology
• a state of the world consistent with the axioms

• Problem we get very large models
• usually results in memory exhaustion

11
Problem 2 Many Models

• OWL axioms are nondeterministic

Each heart is an organ ? Each thing is either
not a heart or it is an organ

• Requires guessing

• Problem there are many combinations of guesses

12
HermiT Addresses These Two Problems

• Based on hypertableau
• well-known from resolution reasoning
• Two optimizations
• guessing reduced to necessary minimum
• model sizes reduced through anywhere blocking
• Web site
• http//web.comlab.ox.ac.uk/oucl/work/boris.motik/H
  ermiT/

13
Performance Evaluation (I)

• GALEN original before simplifications were made
  some 10 years ago
• very hard ontology
• only HermiT can handle it

14
Performance Evaluation (II)

• 83 of the 88 ontologies in the OBO Foundry are in
  Horn-SHIQ

100,000
79 ontologies
10,000
80 ontologies
1,000
100
10
Total Time (s)
82 ontologies
1
0.1
0.01
0.001
Number of Classified Ontologies