Download presentation
Presentation is loading. Please wait.
Published byJasper French Modified over 8 years ago
1
Benchmarking Matching Applications on the Semantic Web
2
Authors Alfio Ferrara & Stefano Montanelli Universita` degli Studi di Milano, DICo - via Comelico 39, 20135 Milano, Italy {ferrara,montanelli}@dico.unimi.it Jan Noessner & Heiner Stuckenschmidt KR & KM Research Group University of Mannheim, B6 26, 68159 Mannheim, Germany {jan,heiner}@informatik.uni-mannheim.de
3
The Semanic Web: Research and Applications 8th Extended Semantic Web Conference, ESWC 2011, Heraklion, Crete, Greece, May 29 – June 2, 2011, Proceedings, Part II Editors: Grigoris Antoniou, Marko Grobelnik, Elena Simperl, Bijan Parsia, Dimitris Plexousakis, Pieter De Leenheer, Jeff Pan ISBN: 978-3-642-21063-1 (Print) 978-3-642-21064-8 (Online)
4
Problem background In the literature most of the existing approaches/techniques use their individually created benchmarks for evaluation In the fields of object reconciliation, duplicate detection, and entity resolution a widely used set of benchmarks are proposed by the Transaction Processing Performance Council (TPC) that focuses on evaluating transaction processing and databases A number of benchmarks are also available for XML data management
5
Problem background not defined in a Semantic Web language, terminological complexity is usually very shallow OAEI organizes since 2005 an annual campaigns 2009 OAEI benchmark: basic level of flexibility and the limited size of test cases benchmark provided by Yatskevich et al., using the taxonomy of google and yahoo as input: the problem to create an error-free gold standard ……
6
The SWING approach Data acquisition techniques support defining a subset of data by choosing both the data categories of interest and the desired size of the benchmark we add semantics to the data acquired Data transformation techniques TBox unchanged, ABox modified in several ways, called test cases data referred to the same objects are provided in different datasources a number of datasets with a variable level of data quality and complexity Data evaluation techniques create a ground-truth as a reference alignment for each test case A reference alignment contains the mappings between the reference ABox individuals and the corresponding transformed individuals in the test case
7
The SWING approach
8
Data Acquisition data selection performed according to an initial query that is executed against a linked data repository with the supervision of the evaluation designer the size of the linked data source is narrowed down by i) selecting a specific subset of all available linked data classes and ii) limit the individuals belonging to these selected classes. data enrichment
9
Data Acquisition
10
data selection data enrichment Add Super Classes and Super Properties bottom-up & top-down Convert Attributes to Class Assertions Determine Disjointness Restrictions Enrich with Inverse Properties Specify Domain and Range Restrictions
11
Data Transformation Goal : Input : a reference ontology and a configuration scheme C Output : a new ontology, and denote two ABoxes consistent with the TBox of the ontology
12
Data Transformation Procedure Preprocessing of the Initial Ontology take into account all the data properties ∈ O and, for each property, we add a new object property add to O new property such that Deletion/Addition of Individuals Individuals Transformation
13
Data Transformation Procedure Preprocessing of the Initial Ontology Deletion/Addition of Individuals Goal : to avoid that those matching applications that produce only one-to- one mappings are favored given and deterministic and non-deterministic strategies Individuals Transformation
14
Data Transformation Procedure Preprocessing of the Initial Ontology Deletion/Addition of Individuals Individuals Transformation Data value transformation Data structure transformation Data semantic transformation
15
Data Individuals Transformation Procedure
16
Data value transformation
17
Data Individuals Transformation Procedure Data structure transformation
18
Data Individuals Transformation Procedure Data semantic transformation
19
Experimental Result simple matching TermMatch StringMatch TextMatch complex matching LexicalMatch StructuralMatch HMatch
20
Experimental Result
22
Thank you for watching ! 王子威 121220099
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.