Presentation is loading. Please wait.

Presentation is loading. Please wait.

IRDQL – Imprecise RDQL Queries Using Similarity Joins Abraham Bernstein, Christoph Kiefer Banff, February 11, 2014.

Similar presentations


Presentation on theme: "IRDQL – Imprecise RDQL Queries Using Similarity Joins Abraham Bernstein, Christoph Kiefer Banff, February 11, 2014."— Presentation transcript:

1 iRDQL – Imprecise RDQL Queries Using Similarity Joins Abraham Bernstein, Christoph Kiefer Banff, February 11, 2014

2 iRDQL – Imprecise RDQL Queries Using Similarity Joins 2 Surfing the Beach Questions: Where to rent a surf board? Which beach service offers best prices? Where to find the biggest waves?

3 iRDQL – Imprecise RDQL Queries Using Similarity Joins 3 RDQL – RDF Data Query Language Jena – Implementing the Semantic Web Recommendations [Carroll et al. 03] SELECT ?s1,?p1 WHERE (?s1 presents ?p1) (?p1 serviceName beach surfing) ?s1?p1 Beach Surfing Service Beach Surfing Profile

4 iRDQL – Imprecise RDQL Queries Using Similarity Joins 4 iRDLQ – Extending RDQL With Similarity Joins 3 additional language constructs IMPRECISE, SIMMEASURE, OPTIONS SELECT ?s1,?p1,?p2 WHERE (?s1 presents ?p1) (?p2 serviceName beach surfing) IMPRECISE ?p1,?p2 SIMMEASURE Levenshtein OPTIONS IGNORECASE false THRESHOLD 0.7 ?s1?p1?p2sim Beach Surfing Service Beach Surfing Profile 1.0 Beach Broker ServiceBeach Broker Profile Beach Surfing Profile 0.85 Abstract Broker Service Abstract Broker Profile Beach Surfing Profile 0.7 Similarity Join

5 iRDQL – Imprecise RDQL Queries Using Similarity Joins 5 Similarity Measurement – Status Quo Similarity between feature vectors [Lee et al. 98] –features of objects like name, degree,... –Cosine of angle strings or sequences of strings [Levenshtein 66] –textual description of objects –Levenshtein, TFIDF trees and graphs [Shasha et al. 02] –tree/graph comparison –Isomorphisms, Tree-edit distance objects [Resnik 95] –amount of information contained in objects CS Dept US UnderGrad Courses Grad Courses People FacultyStaff Associate Prof Prof name: Mike Meyers granting-institution: NYU CS Dept Swiss Courses Staff Academic Staff Technical Staff Lecturer Professor first-name: Abraham last-name: Bernstein degree: Prof., Ph.D. Administration Staff This is the Department of Informatics at the University of Zurich. Computer Science Department at the University of NY. p =

6 iRDQL – Imprecise RDQL Queries Using Similarity Joins 6 Evaluation Approach Quantitative evaluation using an OWL-S service retrieval test collection [Klusch 05] OWL-S Service-based Precision, Recall and F-Measure as performance measures Precision / Recall / F-Measure:

7 iRDQL – Imprecise RDQL Queries Using Similarity Joins 7 OWL-S Service Retrieval Test Collection hotel reservation booking service Provide the best hotel reservation system in a given city. city_broker_service.owls city_broker_service2.owls city_financial_agent_service.owls city_financial_agent_service1.owls urbanarea_financial_agent_service.owls city_organization_service.owls OWL-S services of 6 different domains 9 queries together with its correct answers Query Relevance Set Query

8 iRDQL – Imprecise RDQL Queries Using Similarity Joins 8 iRDQL – Performance (iP vs. iPM) iP: IMPRECISE ?p1, ?p2 iPM: IMPRECISE ?p1, ?p2 & IMPRECISE ?m1, ?m2 Average Precision, Recall, and F-Measure (iP vs. iPM), Levenshtein avg precision (iPM) avg precision (iP) avg recall (iPM) avg recall (iP) avg f-measure (iPM) avg f-measure (iP)

9 iRDQL – Imprecise RDQL Queries Using Similarity Joins 9 iRDQL – Performance (iRDQL vs. OWLS- M4) iRDQL slightly outperformed by specialized algorithm OWLS-M4: Matchmaking algorithm of OWLS- MX [Klusch et al. 05] Precision Recall F-Measure

10 iRDQL – Imprecise RDQL Queries Using Similarity Joins 10 Conclusions and Outlook Combination of RDQL and similarity measures Generic IR-based approach only slightly outperformed Inspired by Cohens work [Cohen 00] no flat tables aggregated (ontologized) objects Performance improvements Switch to SPARQL Thank You...


Download ppt "IRDQL – Imprecise RDQL Queries Using Similarity Joins Abraham Bernstein, Christoph Kiefer Banff, February 11, 2014."

Similar presentations


Ads by Google