Evolution in OWL 2 QL OWL 2 EL

Evolution in OWL 2 QL & OWL 2 EL Ontologies Dmitriy Zheleznyakov 28 th of January, 2014, Oslo

Ontology General rules: o To use ontologies in applications, we need special, formal syntax All popes are clerics Facts: Benedict XVI is a pope 2

Ontology o Do ontologies differ from data bases? o Data bases: explicit knowledge only o Benedict XVI is a pope o Ontologies: explicit & implicit knowledge o Benedict XVI is a pope o Reasoning: Benedict XVI is a cleric reasoning Explicit knowledge Implicit knowledge 2

Ontology Languages o The focus of this work: ontology languages for the Semantic Web o Web Ontology Language: OWL 2 (W 3 C Standard) o OWL 2 QL o OWL 2 EL o Good computational properties o Efficient schema and data management o Used in practice 3

OWL 2 QL: Ontology-Based Data Access o Ontology-Based Data Access (OBDA) o provide unified query interface to heterogeneous data sources 4

OWL 2 QL: Ontology-Based Data Access o Ontology-Based Data Access (OBDA) o provide unified query interface to heterogeneous data sources o EU FP 7 project Optique will develop an OBDA system o use-case partners: Statoil, Siemens o Ontologies may change: o new knowledge about domain o new data source is added o Motivation for our work: o to address the dynamicity of OBDA systems by studying evolution of schema and data 4

OWL 2 EL: Clinical Science, Bio Ontologies enable communication and knowledge sharing between doctors, scientists, etc. o SNOMED CT: > 311 k terms o constantly under development: o 5 modification teams o every 2 weeks the main team integrates changes, o 2002 2008 SNOMED went 278 k 311 k terms o It is the standard to describe the results of experiments in the US clinical labs o Motivation for our work: o to provide techniques that facilitate ontology development for such a vast community 5

Our Goal o To facilitate evolution of ontology-based systems o insertion of knowledge o deletion of knowledge o On two levels: o schema o data Original ontology To insert o With as little changes as possible To delete 6

How to Approach the Problem? Original ontology 1. Define an operator and understand it New knowledge • a conceptual understanding of how to evolve ontologies • checking its computational properties 2. Develop an algorithm to compute the result 3. Implement the algorithm Resulting ontology 7

Previous Work Model-based operators Formula-based operators Many evolution operators proposed [AGM’ 85] [Borgida’ 85] [Dalal’ 88] [Winslett’ 88] [Satoh’ 88] [Katsuno&Mendelzon’ 91] [Winslett’ 90] AI: 80’s – 90’s Propositional logic, weaker then OWL 2 QL & OWL 2 EL Adaptation of some operators [Kang&Lau’ 04] [Liu& al’ 06] [Flouris&al’ 04] [Qi&Du’ 09] [Flouris&al’ 05] [De. Giacomo&al’ 07 -09] [Qi&al’ 06] [Wang&al’ 10] KR: 2004 -2006 2007 -2010 8

General Overview of the Results Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 For OWL 2 QL & EL - inexpressibility OWL 2 QL - counterintuitive results OWL 2 EL - inexpressibility - counterintuitive results 9

Understanding Model-Based Operators Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 10

Understanding Model-Based Operators o We have shown: operators are determined by three parameters o this gives a three-dimensional space of operators o Classical operators fit in this space o Novel operators can be easily defined by changing parameters 11

Understanding Model-Based Operators o We noticed: operators are determined by three parameters o this gives a three-dimensional space of operators o Classical operators fit in this space o Novel operators can be easily defined by changing parameters o We can add new values to dimensions! o more operators can be defined! 11

Inexpressibility of Model-Based Operators Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 - inexpressibility - counterintuitive results 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 12

Inexpressibility of Model-Based Operators Schema: Wives are married to their husbands Priest cannot be husbands Facts: Mary is married to John and Adam and Bob are priests 13

Inexpressibility of Model-Based Operators a model: Priest Adam has. Husband Mary John Bob Under model-based operators: We incorporate new knowledge directly into models Facts to add: John is a priest 13

Inexpressibility of Model-Based Operators Priest Adam has. Husband Mary 1. has. Husband Adam John Bob John cannot be a husband of Mary anymore! What happens to her? Priest John 2. Three options: 1. She divorced 2. She married some one else 3. She married to a former priest Priest Adam has. Husband Mary Jack Bob John 3. Priest Adam has. Husband Mary Bob John 13

Inexpressibility of Model-Based Operators Priest Adam has. Husband Mary 1. has. Husband Adam John Bob We showed: all these options cannot be captured in OWL 2 QL and OWL 2 EL Priest John OR 2. Priest Adam We need at least disjunction which is not in OWL 2 QL and OWL 2 EL has. Husband Mary Jack Bob John OR 3. Priest Adam has. Husband Mary Bob John 13

Bad Behaviour of Model-Based Operators Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 - inexpressibility - counterintuitive results 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 14

Bad Behaviour of Model-Based Operators Some of model-based operators behave as follows: No schema Facts: Adam and Bob are priests Facts to add: John is a priest 15

Bad Behaviour of Model-Based Operators Some of model-based operators behave as follows: Priest Adam Bob Such behaviour is not useful for any application Expected result: Priest Adam Actual result: Priest John Bob John 15

Restriction of OWL 2 QL Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 16

Restriction of OWL 2 QL o We found the reason of the bad behaviour of model-base operators: A binary relation participates in disjointness o Priest cannot be husbands o What if we forbid this bad interaction? o We showed: most of model-based operators work! o this fragment captures (FO part of) RDFS (another W 3 C standard) Priest Adam has. Husband Mary John Bob disjoint with 17

Summing up on Model-Based Operators o Model-based operators o suffer from inexpressibility o tend to lose too much of information o counterintuitive behaviour o Our verdict: o model-based operators are not suitable for the case of OWL 2 QL or OWL 2 EL o We turned to Formula-based operators! 18

Formula-Based Operators Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 - inexpressibility - counterintuitive results 19

Formula-Based Operators o Preserve all the knowledge: both explicit and implicit o Example: delete Priests are Males o We do not want to lose info that Adam is Male Explicit schema Explicit data Implicit schema Implicit data 20

Formula-Based Operators o Preserve all the knowledge: both explicit and implicit o Example: delete Priests are Males Explicit schema Implicit schema o How to delete it in such a way that it will not appear even implicitly? o Delete o either Priests are Clerics o or Clerics are Males 20

Formula-Based Operators o Preserve all the knowledge: both explicit and implicit The resulted schema: either or o Example: delete Priests are Males o How to delete it in such a way that it will not appear even implicitly? o Delete o either Priests are Clerics o or Clerics are Males o What to do with a multiple choice? Classical approaches: o Keeping both – impossible o Combining them o too much of information is lost o we proved: it is computationally hard 20

Bold Operator Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 21

Bold Operator There is no way to decide o which result is delete better! Priests are Males Example: This is application dependent and o How to delete it in such a way should be up to the user that it will not appear even implicitly? o Delete o either Priests are Clerics The resulted schema: o or Clerics are Males either or o What to do with a multiple choice? o We propose: Bold operator. It picks up one of them o The result is non-deterministic… o … But can be computed in polynomial time (for OWL 2 QL) o In the case of OWL 2 EL: o Implicit knowledge can be infinite o Bold operator does not work 22

Tunable Operator Work for restriction of OWL 2 QL Model-based operators Formula-based operators Propositional logic 1 4 OWL 2 QL OWL 2 EL 2 3 Bold operator Works for OWL 2 QL Tunable operator Works for OWL 2 QL & EL 5 6 23

Tunable Operator o Tunable operator o allows to choose what part of implicit knowledge will be preserved Explicit schema Implicit schema 24

Tunable Operator o Tunable operator o allows to choose what part of implicit knowledge will be preserved Explicit schema Implicit schema 24

Tunable Operator No implicit part Whole implicit part 25

Summing up on Formula-Based Operators o Classical Formula-based operators o suffer from inexpressibility o tend to lose too much of information o Bold operator: o works for OWL 2 QL o fails for OWL 2 EL o Tunable operator: o works for both OWL 2 QL and OWL 2 EL 26

Current Work o Applying our results to the Optique project o in progress o Incorporating evolution in transition systems o IJCAI’ 2013 o Information hiding & Controlled query evaluation o ISWC 2013 o submitted to an international conference 27