S-Coordinate: a Calculation and an Execution of Semantic Coordinating.


81 views
Uploaded on:
Category: Product / Service
Description
first European Semantic Web Symposium, 11 May 2004, Crete, Greece ... first European Semantic Web Symposium, 11 May 2004, Crete, Greece. For all marks in T1 and T2 register ideas ...
Transcripts
Slide 1

S-Match: an Algorithm and an Implementation of Semantic Matching Pavel Shvaiko paper with Fausto Giunchiglia and Mikalai Yatskevich 1 st European Semantic Web Symposium, 11 May 2004, Crete, Greece

Slide 2

Semantic Matching The S-Match Algorithm The S-Match System: Architecture and Implementation A Comparative Evaluation Future Work Outline

Slide 3

Semantic Matching

Slide 4

Matching: given two chart like structures (e.g., idea progressions or ontologies), create a mapping between the hubs of the diagrams that semantically relate to each other Matching Syntactic Matching Semantic Matching Relations are figured between ideas at hubs R = { = , , } Matching Relations are processed between names at hubs R = {x  [0,1]} Note: First usage CTXmatch [Bouquet et al. 2003 ] Note: every single past framework are syntactic…

Slide 5

Mapping component is a 4-tuple < ID ij , n 1 i , n 2 j , R > , where ID ij is an exceptional identifier of the given mapping component; n 1 i is the i-th hub of the principal diagram; n 2 j is the j-th hub of the second chart; R indicates a semantic connection between the ideas at the given hubs Computed R\'s , recorded in the diminishing restricting quality request: proportionality { = }; more general/particular { , }; mismatch {  }; overlapping { }. Semantic Matching Semantic Matching: Given two charts G1 and G2 , for any hub n 1 i  G1 , locate the most grounded semantic connection R\' holding with hub n 2 j  G2

Slide 6

? Pictures Europe = ? Europe Pictures Wine and Cheese ? Austria Italy Austria < ID 22 , Europe , Pictures , = >  < ID 21 , Europe , Europe , > < ID 22 , Europe , Pictures , = > < ID 24 , Europe , Italy , > Example: Two straightforward idea progressive systems Algo Step 4

Slide 7

The S-Match Algorithm

Slide 8

For all names in T1 and T2 register ideas at names For all hubs in T1 and T2 process ideas at hubs For all sets of names in T1 and T2 figure relations between ideas at marks For all sets of hubs in T1 and T2 process relations between ideas at hubs Steps 1 and 2 constitute the preprocessing stage, and are executed once and every time after the outline/metaphysics is changed (OFF-LINE part) Steps 3 and 4 constitute the coordinating stage, and are executed each time the two patterns/ontologies are to be coordinated (ON - LINE section) Four Macro Steps Given two named trees T1 and T2, do:

Slide 9

The thought: Translate regular dialect expressions into interior formal dialect Compute ideas taking into account conceivable faculties of words in a name and their interrelations Preprocessing: Tokenization. Marks (as per accentuation, spaces, and so on.) are parsed into tokens. E.g., Wine and Cheese  <Wine, and, Cheese> ; Lemmatization . Tokens are morphologically examined with a specific end goal to discover all their conceivable fundamental structures. E.g., Images  Image ; Building nuclear ideas. A prophet (WordNet) is utilized to concentrate faculties of lemmatized tokens. E.g., Image has 8 detects, 7 as a thing and 1 as a verb ; Building complex ideas . Relational words, conjunctions, and so on are interpreted into sensible connectives and used to assemble complex ideas out of the nuclear ideas E.g., C Wine and Cheese = <Wine, U(WN Wine )> <Cheese, U(WN Cheese )> Step 1: register ideas at names

Slide 10

Europe 1 Pictures Wine and Cheese 2 3 4 5 Austria Italy C Italy = C Europe C Pictures C Italy Step 2: c ompute ideas at hubs The thought: broaden ideas at names by catching the information living in a structure of a diagram with a specific end goal to characterize a setting in which the given idea at a mark happens Computation: Concept at a hub for some hub n is processed as a crossing point of ideas at names situated over the given hub, including the hub itself

Slide 11

The thought: Exploit from the earlier learning, e.g., lexical, space information Strong semantics component level matchers. Separate semantic relations utilizing prophets (WordNet): Equivalence: An is equal to B , iff there is no less than 1 sense in A which is an equivalent word of a sense in B ; More broad: An is more broad than B iff there is no less than 1 sense in A that has a sense in B as hyponym or meronym; Less broad: An is less broad than B iff there is no less than 1 sense in A that has a sense in B as hypernym or holonym; Mismatch: A befuddles with B if there are two detects (one from every) which are diverse hyponyms of the same synset or in the event that they are antonyms. Frail semantics component level matchers. String-based, sense-based, and so forth.: Prefix: net is thought to be proportionate to arrange ; Expansion: P.O. is thought to be equal to Post Office ; Soundex: Fausto is thought to be comparable to Phausto. Step 3: process relations between ideas at names

Slide 12

T1 T2 Europe Images 1 Wine and Cheese Pictures Europe 2 3 T1 T2 C Europe C Pictures C Italy C Austria Italy C Wine C Cheese 4 5 Austria Italy 3 4 C Images = C Europe = C Austria  = C Italy =  Step 3: cont\'d Recall the illustration: Results of step 3:

Slide 13

Context  rel ( C1 i , C2 j ) A propositional recipe is legitimate iff its invalidation is unsatisfiable SAT deciders are sound and finish … Construct the propositional equation C1 i = C2 j is deciphered into C1 i  C2 j C1 i C2 j is interpreted into C1 i  C2 j (comparably for ) C1 i  C2 j is interpreted into ¬ ( C1 i  C2 j ) rel = { = , , }. Step 4: register relations between ideas at hubs The thought: Reduce the coordinating issue to a legitimacy issue We take the relations between ideas at names processed in step 3 as maxims ( Context ) for thinking about rel ations between ideas at hubs .

Slide 14

C2 Pictures C1 Europe Context If a check for < ,  > comes up short, then covering is returned T1 T2 C Europe C Pictures C Wine and Cheese C Italy C Austria C Images C Europe = C Austria  C Italy =  Step 4: cont\'d Example. Assume we need to check if C1 Europe = C2 Pictures (C1 Images  C2 Pictures )  (C1 Europe  C2 Europe )  (C1 Images  C1 Europe )  (C2 Europe  C2 Pictures )

Slide 15

T1 T2 Europe Images 1 Wine and Cheese Wine and Cheese Wine and Cheese Wine and Cheese Pictures Europe 2 3 Italy 4 5 Austria Italy 3 4 Step 4: cont\'d = 

Slide 16

The S-Match System: Architecture and Implementation

Slide 17

S-Match: Logical Level NOTE: Current variant of S-Match is an excused re-execution of the CTXmatch framework with a couple included functionalities

Slide 18

S-Match: Algorithmic Level Off-line part (Steps 1,2): Java WordNet Library (JWNL) 1.3 WN 2.0 (content record or database or memory occupant database) On-line part (Steps 3,4): Strong semantics matchers WordNet 2.0 Weak semantics matchers (12) String-based Sense-based Two SAT solvers (JSAT, SAT4J)

Slide 19

A Comparative Evaluation

Slide 20

Testing Methodology Matching frameworks S-Match versus Cupid, COMA and SF as actualized in Rondo Measuring match quality Expert mappings are characteristically subjective Two degrees of opportunity Directionality Use of Oracles Indicators Precision, [0,1] Recall, [0,1] Overall, [-1,1] F-measure, [0,1] Time, sec.

Slide 21

Three investigations, experiments from various areas Some qualities of experiments: #nodes 4-39, profundity 2-3 Preliminary Experimental Results PC: PIV 1,7Ghz; 256Mb. RAM; Win XP

Slide 22

Extend the semantic coordinating way to deal with permit taking care of diagrams Extend the semantic coordinating calculation for figuring mappings between charts Develop a hypothesis of iterative semantic coordinating Elaborate results sifting methodologies as indicated by the coupling quality of the subsequent mappings Optimize the calculation and its usage Develop GUI to make the framework intuitive Extend libraries Develop semantic coordinating testing technique Do throught testing of the framework Future Work

Slide 23

Project site - ACCORD: http://www.dit.unitn.it/~accord/F. Giunchiglia, P.Shvaiko, M. Yatskevich: S-Match: a calculation and an execution of semantic coordinating . In Proceedings of ESWS\'04 . F. Giunchiglia, P.Shvaiko: Semantic coordinating . To show up in The Knowledge Engineering Review diary, 18(3) 2004. Short forms in Proceedings of SI workshop at ISWC\'03 and ODS workshop at IJCAI\'03 . P. Bundle, L. Serafini, S. Zanobini: Semantic coordination: another methodology and an application . In Proceedings of ISWC\'03 . F. Giunchiglia, I. Zaihrayeu: Making peer databases communicate – a dream for an engineering supporting information coordination . In Proceedings of CIA\'02. C. Ghidini, F. Giunchiglia: Local models semantics, or logical thinking = territory + similarity . Manmade brainpower diary, 127(3):221-259, 2001. References

Slide 24

Thank you!

Slide 25

System Matches Expert Matches B A C D A – False negatives B – True positives C – False positives D – True negatives

Recommended
View more...