Deriving an Emergent Relational Schema from RDF Data
0 likes874 views
This document discusses deriving an emergent relational schema from RDF data. It describes extracting characteristic sets from RDF data to recognize classes and relationships between classes. These characteristic sets are then merged and labeled to create a logical relational schema. This emergent schema provides benefits for both systems through improved efficiency and humans through easier query formulation over the RDF data. Key aspects of a useful emergent schema are discussed such as being compact, having human-friendly labels, providing high coverage of the RDF data, and being efficient to compute. Experimental results on real-world RDF datasets show the approach produces compact schemas with high coverage and understandable labels that improve performance over the native RDF representation.
![ Bad query plans
Low storage locality
Lack of user schema insight
Main Problems in RDF Data Management [1]
[1] Minh-Duc Pham, Boncz P.A., " Self-organizing Structured RDF in MonetDB ," PhD Symposium, ICDE, 2013
RDF de-emphasizes the need for a schema
and the notion of structure in the data
Emergent Schema](https://image.slidesharecdn.com/emergentschemagraphta-150323090108-conversion-gate01/85/Deriving-an-Emergent-Relational-Schema-from-RDF-Data-2-320.jpg)
Deriving an Emergent Relational Schema from RDF Data
- 1. Deriving an Emergent Relational Schema from RDF Data Minh-Duc Pham Linnea Passing Orri Erling Peter Boncz duc@cwi.nl linnea.passing@tum.de oerling@openlinksw.com boncz@cwi.nl
- 2. Bad query plans Low storage locality Lack of user schema insight Main Problems in RDF Data Management [1] [1] Minh-Duc Pham, Boncz P.A., " Self-organizing Structured RDF in MonetDB ," PhD Symposium, ICDE, 2013 RDF de-emphasizes the need for a schema and the notion of structure in the data Emergent Schema
- 3. Emergent schema = “rough” schema to which the majority of triples conforms Recognize: Classes (Characteristic Sets - CS’s ) – recognize “classes” of often co-occurring properties Relationships (CS) – recognize often-occurring references between such classes + give logical names to these <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://rdfs.org/sioc/ns#num_replies> <http://purl.org/dc/terms/title> <http://rdfs.org/sioc/ns#has_creator> <http://purl.org/dc/terms/date> <http://purl.org/dc/terms/created> <http://purl.org/rss/1.0/modules/content/encoded> <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://xmlns.com/foaf/0.1/name> <http://xmlns.com/foaf/0.1/page> <has_creator> “Book” Recovering the Emergent Schema of RDF data “Author”
- 4. inproc1 creator inproceeding author4 inproc2 type inproceeding author2 “BBB” inproc3 inproceeding author3 “CCC” conf1 conf2 Conference “conference1” “2010” webpage1 ID type creator title partOf inproc1 inproceeding {author3, author4} “AAA” conf1 inproc2 inproceeding author2 “BBB” conf1 inproc3 inproceeding author3 “CCC” conf2 ID type title issued conf1 Conference “conference1” 2010 conf2 Proceedings “conference2” 2011 Foreign Key Relationship Irregularity conf2 “content.php” “Table Cont.” “index.php” “index.php” “content.php” Original RDF graph Self-organizing RDF representation Example of structure recognized from RDF graph
- 5. Relational Schema Semantic Web Schema Describes the structure of the occurring data Purpose: knowledge representation Concept mixing (for convenience) Describing a concept universe (regardless data) Designed for one database (=dataset) Designed for interoperability in many contexts Statement: it is useful to have both an (Emergent) Relational and Semantic Schema for RDF data useful for systems (higher efficiency) useful for humans (easier query formulation) What does “schema” mean?
- 6. Compact Schema • as few tables as possible • homogeneous literal types (few NULLs in the tables) Human-friendly “Labels” • URIs + human-understandable table/column/relationship names High “Coverage” • the schema should match almost all triples in the dataset Efficient to compute • as fast as data import When is a Emergent Schema of RDF data useful?
- 8. Characteristic Sets in some well-known RDF datasets
- 9. Partial and Mixed Use of Ontologies
- 10. Emerging a Relational Schema CS2 CS2 S P O ... 1. Extract basic CS’s 4. Schema Filtering Triple table 3. Merge similar CS’s 5. Instance Filtering Physical relational schema Basic CS’s CS4 CS0CS5 CS2 CS3 CS1 label1 label4 label5 Label3 CS4 CS0CS5 CS2 CS3 CS5 Merged CS’s CS0CS5 CS7 CS6 CS6 CS7 CS5 Merged CS’s CS0CS5 CS7 CS6 2. Labeling Parameter Tuning
- 11. Results: compact schemas with high coverage
- 12. Results: understandable labels & performance Likert Score: 1=bad ….. 5=excellent