Dynamic Ontology Matching Pavel Shvaiko OpenKnowledge meetings 9 February, 13 March, 2006 Trento, Italy.

Dynamic Ontology MatchingDynamic Ontology Matching

Pavel Shvaiko

OpenKnowledge meetings

9 February, 13 March, 2006

Trento, Italy

2

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Introduction (Trento view)Information sources (e.g., catalogs) can be viewed as graph-like structures containing terms and their inter-relationships

Matching takes two graph-like structures and produces a mapping between the nodes of the graphs that correspond semantically to each other

3

OK meetings, 9 February, 13 March, 2006, Trento, Italy

P2P scenario (match-oriented view)

4

OK meetings, 9 February, 13 March, 2006, Trento, Italy

P2P scenario: more details

Peers are autonomous

• They appear and disappear on the network

• They use different terminology

Matching (on-the-fly)

• Determine the relationships between peer schemas

• Use these relationships for query answering

• An assumption that all peers rely on one global schema, as in data integration, can not be made, because the global schema might need to be updated any time the system evolves

5

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Requirements

Input size of ontologies

At most 100 entinties per ontology

Domains of interest

Bioinformatics and GIS Emergency response

Matching Performance

At most 2 seconds per matching task

Memory limit: 256Mb

Matching Quality Mistakes are acceptable

6

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Discussion - IInput

OWL, RDF, XML

Will the instances be available?

Quality/charachteristics of entities

Partial vs Complete ontology matching

Perhaps we might not need to have a complete alignment to answer a

query

Quality/Efficiency trade off

QOM example

Online vs Offline vs Mixed match and QA

7

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Discussion - II

What is in the alignment ?

1-1, 1-n, n-m

Is any relation suitable?

Output format

Test cases

The sooner we have them, the better

Matching quality measures

User/task related measures

What is more important in the application:

Precision or recall or both?

8

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Discussion - III

Alignment negotiation

Explanation and argumentation

9

OK meetings, 9 February, 13 March, 2006, Trento, Italy

A comparison of techniques for dynamic ontology matching

1. Introduction [2p] All

2. The dynamic ontology matching problem [19p] Pavel1. P2P information management systems [3p] Ilya2. Motivating scenarios (2 our applications) [12p] Maurizio+Marco + Marta?

3. Requirements (functional vs non-functional) [2p]Pavel + Marta?

4. Problem statement [2p] Pavel+Ilya1. why is it different from previous works

3. A conceptual basis for comparison of dynamic matching techniques [13p] Marco

1. The framework + taxonomy [4+3p] Marco+Pavel+Mikalai

2. Ontology matching (standard) [3p]Pavel+Mikalai

3. Plausible DOM methods (transitivity) [3p]Pavel+Mikalai

4. Systems and evaluation [10p] Mikalai1. State of the art prototypes [2p]Pavel

2. Evaluation methodology [3p]Mikalai

3. Comarative evaluation results [5p]Mikalai

5. Discussion/Open Issues and challenges towards DOM [3p] All

6. Conclusions [2p]

10

OK meetings, 9 February, 13 March, 2006, Trento, Italy

A comparison of techniques for dynamic ontology matching

Index solid: Feb 17 (DONE)

Parallel 2,3,4: March 10 (DONE)

Use case + details of what should be matched by Fiona (March 22)

A first draft (circulated to partners): April 5

Trento

Feedback by April 19

Second draft (circulated to parnters):

Barcelona

Feedback by

Final version by mid May?

Trento

11

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Motivating scenarios (P2P + 2 our applications)

1. Intuitive description (environment, actors, operations for the system) [1p]

2. Requirements (Tropos) [2p + 1 fig]domain description (Peers and peer goals)

use case

QA (functional requirements)1. Measure quality (GEA) Trade quality for speed?

2. Transitivity in GEA

3. Logical architecture (organization of users and C/S Ps) [1p+1fig]

4. Physical architecture (bioinformatics=logical arcitechture)

5. Non-functional requirements [1p] P2P1. Number of peers and connectivity

2. Size and shape of ontologies/data

3. Run-time vs offline: time response, mixed initiative

4. Memory limit (256 mb)

12

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Conceptual Framework (Marco->March) [m6]1. Introduction

2. P2P … 1. P2P information management systems

2. Motivating examples

3. Basic notation, terminology

4. Ontology Matching1. Running examples (semantic matching + IF-MAP)

5. DOM1. Dynamics (peers, ontologies, …)

2. Transitivity (compositionality of mappings and queries)

3. The basic theorem

6. DOM interaction model

7. Formalizing motivating examples

13

OK meetings, 9 February, 13 March, 2006, Trento, Italy

Methodological Framework (Trento->March)1. Routing / Navigation / Search Ilya + Maxym [m6]

1. Basic operations 1. Node matching

2. Navigation

3. Query answering (substeps: rewriting)

2. Composite operations1. Matching

2. QA

3. Interaction models for the above 2

4. Two case studies

2. Approximation / Quality [m12]1. ???

14

OK meetings, 9 February, 13 March, 2006, Trento, Italy

A potential example of DOM - 1

[Source: Bin He]

15

OK meetings, 9 February, 13 March, 2006, Trento, Italy

A potential example of DOM - 2

[Source: Bin He]

16

OK meetings, 9 February, 13 March, 2006, Trento, Italy

DOM: open questions

1. What do we technically mean by dynamic? ontology matching

2. Business cases & technical use cases

3. Technically, what do we match in our scenarios?1. Messages between agents

2. Functionalities of web services

3. Classifications/Ontologies