30.10.06 Krzysztof Janowicz SIM-DL -Towards a Semantic Similarity Measurement Theory for the Description Logic ALCNR in Geographic Information Retrieval.

•

30.10.06 Krzysztof Janowicz

SIM-DL-Towards a Semantic Similarity Measurement Theory for the Description Logic ALCNR in Geographic Information Retrieval

Krzysztof JanowiczInstitute for Geoinformatics; University of Münster

Krzysztof Janowicz SIM-DL (for ALCNR) 2

Outline

• Motivation: Yet Another Similarity Theory?

• Similarity & Subsumption based IR

• Matching Scenario

• SIM-DL Framework

• Human Subject Testing

• Results, Conclusions & Outlook

Krzysztof Janowicz SIM-DL (for ALCNR) 3

Yet Another Similarity Theory?

Available ontologies (DL!)Available theories

Measured between (re)representation!

http://flickr.com/photos/genista/25390358/

Krzysztof Janowicz SIM-DL (for ALCNR) 4

Similarity & Subsumption based Retrieval

Krzysztof Janowicz SIM-DL (for ALCNR) 5

Similarity vs. Subsumption

• Subsumption-based Retrieval(+) Results fit user’s requirements (subconcepts!)(-) Too generic / too specific result set(-) Artificial search concept

• Similarity-based Retrieval(+) Search concept = searched concept(-) Results not necessarily fit user’s requirements

Krzysztof Janowicz SIM-DL (for ALCNR) 6

Matching-Scenario

• Accommodation web portal

• External services (SOA)

• Use shared base vocabulary

• Local interface and terminology

• Hotel, Houseboat, Youth Hostel, Botel,….

Task: Integrate Amsterdam-Accommodation Service

Where to put botels?

Krzysztof Janowicz SIM-DL (for ALCNR) 7

Houseboats, Hotel &Botel

Krzysztof Janowicz SIM-DL (for ALCNR) 8

Some Impressions

Pictures received by email, taken from wikipedia and http://www.hotels.nl/amsterdam/botel/

Krzysztof Janowicz SIM-DL (for ALCNR) 9

SIM-DL: Representation (ALCNR)

Krzysztof Janowicz SIM-DL (for ALCNR) 10

SIM-DL: Framework

1. Specify search concept and context

2. Rephrase concepts to canonical NF

3. Generate alignment matrix

4. Apply sim-functions for selected combinations

5. Derive normalized overall similarity

Krzysztof Janowicz SIM-DL (for ALCNR) 11

SIM-DL: Search Concept & Context

(-) Results not necessarily fit user’s requirements

Define Context

Clcs ≡ Housing

Cs ≡ Botel

Krzysztof Janowicz SIM-DL (for ALCNR) 12

Rephrase Concepts to Canonical NF

• ALCNR Normal Form:

+ Rewriting rules (e.g. R() ≡ (≤ 0 R))

+ Minimal set of descriptions (concepts)

Canonical Normal Form

Krzysztof Janowicz SIM-DL (for ALCNR) 13

Generate Alignment Matrix

• Cartesian Product Cs Ct

CsiCsj

Csk

Cti N … …

Ctj … … H

Ctk … CO …

Ctl … … …

HierarchiesNeighborhoodsCo-Occurrence

H > N > CO

Krzysztof Janowicz SIM-DL (for ALCNR) 14

Apply Similarity-Functions (for selected combinations)

• Individual similarity functions for each DL language constructor:{union, intersection, role-intersection, existential

quantification, value restriction, cardinality}

• For Hierarchies, Neighborhoods, Co-Occurrence

edge_distancemax_distance

Krzysztof Janowicz SIM-DL (for ALCNR) 15

Amalgamated & Normalized Overall Similarity

• Union-Constructor:Weighted sum of similarities on CNF union levelWeightings derived from A-Box, T-Box or A&T-Box

• Intersection-Constructor:Sum of similarities on CNF intersection levelNormalization to [0,1]

Krzysztof Janowicz SIM-DL (for ALCNR) 16

Human Subject Testing: Roles & Fillers

Auto. weighted average (>)

Multiplicative approach (<)

User input

Disjoint from watercourse Meets river

Krzysztof Janowicz SIM-DL (for ALCNR) 17

Results, Conclusion & Outlook

• SIM-DL combines subsumption and similarity

• Adapts results from psychology & computer science Cognitive Engineering ;-)

• Only basic model of Alignment and Context

• More Human Subject Tests needed

• More expressive DL

• Usability?

( )ALCRP DALCNRNear ?

•

30.10.06 Krzysztof Janowicz

Questions? Thanks for your attention!

Visit www.similarity-blog.de for related literature.

From: http://www.jobblog.ch/sommer-250