Dave Kolas, BBN Technologies

1
Supporting Spatial Semantics with SPARQL

• Dave Kolas, BBN Technologies
• Terra Cognita 08
• Karlsruhe, Germany 10/26/08

2
Presentation Outline

• Background Information
• What is SPARQL, anyway?
• Why does anyone care about spatial semantics?
• Related Work
• Object-Relational Spatial
• XML Spatial
• Semantic Web Spatial
• Goals of a Spatial Semantic Query Language
• What SPARQL can do now
• What SPARQL cant do now
• Implementation
• Conclusions

3
What is SPARQL, anyway?

• SPARQL is the SPARQL Protocol and RDF Query
  Language
• W3C Candidate Recommendation for Semantic Web
  querying
• RDF Resource Description Framework
• Language for describing data on the web
• Think XML on steroids
• OWL Web Ontology Language
• Language for representing ontologies on the web
• Richer than RDF
• Builds upon RDF as well

4
Who cares about spatial semantics?

• Combining Semantic Web technologies with spatial
  helps with several important problems
• Multiple references to the same place
• Multiple names for the same place
• Allows complex class and property hierarchies and
  relationships for representing types
• Provides a basis for semantic interoperability

5
Related Work Object-Relational Spatial

• Not much work to use an Object-Relational query
  language for querying spatial OR systems
• SQL3 Geometry objects are defined, as well as
  operations that can be performed on them
• This does not provide a transport mechanism for
  the spatial data!

6
Related Work XML Spatial

• XML has been used with spatial data to provide a
  data exchange format and language
• Open Geospatial Consortium GML
• Storing data directly in GML is inefficient,
  difficult to process spatial relationships

7
Related Work SemWeb Spatial

• Relatively new research area
• Provides a higher level of data interoperability
  with other systems, as formal semantics are
  defined
• Provides more complex relationships than XML can

8
Goals of a Semantic Query Language

• Maintain the Graph Theoretical Model of Data
• Do not specialize SPARQL for spatial
• Support appropriate types of spatial queries
• Allow all relevant data to be returned in one
  query

9
Goals Maintain the Graph

• The fundamental data model of the Semantic Web
  languages is a conceptual graph
• This theoretical model holds from end to end
  data is stored as a graph, transferred as a
  graph, and queried as a graph
• Spatial data should not break this paradigm

10
Goals No Specialized SPARQL for Spatial

• A goal of RDF, OWL, and SPARQL is to be domain
  neutral
• No matter what the topic of dicsourse is, you
  should be able to discuss it in this language
• Query clients can be built to create SPARQL
  queries on the fly
• While some clients will obviously know they are
  querying for spatial data, others may not these
  clients should be supported by spatial SPARQL too

11
Goals Types of Queries

• Queries about the spatial properties of an
  individual
• Queries that relate individuals to a known
  location (point and range queries)
• Queries that relate individuals to one another
  (spatial join, nearest neighbor queries)
• Queries that spatially aggregate individuals

12
Goals All Relevant Data in One Query

• One query can be posed for one logical question
• Combinations of the query types above must be
  supported
• Spatial data must be retrievable without knowing
  the geometry types a priori

13
Why not just use functions?

• SPARQL already has a mechanism for function calls
  within a query
• Why not just use that?
• It breaks the graph model (Goal 1) the
  relationships then cannot be serialized in the
  language
• It specializes the language (Goal 2) clients
  have to know they are querying for spatial data

14
What can SPARQL do now?

• SPARQL already has the capacity to do basic
  spatial queries
• We can represent geometries ontologically
• We can represent spatial relationships
  ontologically too
• DESCRIBE can be used to get spatial information
  without knowing geometries

15
Current SPARQL Representing Geometries

• We can use an ontology of geometries so that
  geometry definitions can be put directly in the
  query!
• Base this on GML geometry types, using W3C Geo
  language

a restaurantPizzaParlor restaurantname
Jimmys Pizza georsswhere a
gmlPoint gmlpos 38 -77 .
16
Current SPARQL Representing Relationships

• We can use an ontology of topological
  relationships so that they too can be put in the
  query!
• Base this on some formalism (here, RCC8)

rccdisconnected a owlObjectProperty a
owlSymmetricProperty . rccpart a
owlObjectProperty rdfssubPropertyOf
rccconnected .
We use the terms from RCC8 without necessarily
using the reasoning axioms
17
Current SPARQL Using Describe

• DESCRIBE is a query mechanism whereby a querier
  can request unspecified data about an object
  what is returned is up to the knowledge service
• A spatial knowledge base can use this
  functionality to return geometry infromation,
  whether it is in the form of a point, line,
  polygon, etc.

DESCRIBE ?geo WHERE
lthttp//data/JimmysPizzagt georsswhere ?geo

18
What does SPARQL need?

• There are some things that are awkward or just
  cannot be done with the current specification.
• Range queries are awkward
• Getting back geometry types is challenging
• Count queries are impossible

19
Enhancements Query Premise

• Concept is that querier declares statements to be
  true to be considered with the query
• Query premises have been used in several other
  languages
• KQML
• DAML-QL
• Query premises allow us to cleanly express range
  queries

20
Query Premise Example

• SELECT ?x
• PREMISE
• ?b a gmlBuffer
• gmlradius 1
• gmlbufferGeometry
• a gmlPoint
• gmlpos 38 -77
• .
• WHERE
• ?x a gasGasStation
• georsswhere ?y.
• ?y rccpart ?b .

21
Enhancements Embedded Describe

• The problem is that it is complex to ask for
  things within a spatial region and get their
  geometries in the same query
• Can be done, but needs to use series of OPTIONAL
  clauses (UGLY)
• If we could use the DESCRIBE functionality within
  other types of queries, we could do it!

22
Embedded Describe - Example

• SELECT ?landmark ?geometry DESCRIBE(?geometry)
• WHERE
• ?landmark a lmLandMark
• georsswhere ?geometry.
• ?geometry a ?geometryType
• rccpart ?x.
• va-countiesArlington georsswhere ?x.

23
Enhancements Counting

• Maybe surprisingly, SPARQL cannot count!
• This is due to the lack of a unique names
  assumption
• But we need to count, and sometimes we know we
  can count
• When two things returned must be unique!
• Potential solution count references instead of
  individuals

24
Counting Example

• SELECT COUNT(?x)
• PREMISE
• ?b a gmlBuffer
• gmlradius 1
• gmlbufferGeometry lthttp//data/JimmysLocation
  gt
• WHERE
• ?x a restaurantPizzaParlor
• georsswhere
• rccpart ?b
• .

25
Implementation

• BBN has developed a prototype knowledgebase for
  combining Semantic Web data with spatial data,
  temporal data, and indexing of both
• This KB makes use of the Spatial SPARQL
  suggestions outlined here
• There are also extensions to the Lehigh
  University Benchmark based on this work, for
  testing scalability of spatial Semantic Web
  knowledge bases

26
Conclusions

• SPARQL can already be used for basic querying of
  spatial semantic systems!
• Several enhancements can be added to SPARQL that
  support spatial data while also being generally
  useful
• Query Premises
• Embedded DESCRIBE
• Counting

27
Questions?