134
Omar Boucelma, Mehdi Essid, and Yassine Lassoued
7.2 Related Work
In this section, we describe some related work that will help the reader to understand our approaches to both spatial data integration and geographical data quality
management.
7.2.1 Data Integration
The DataBase (DB) community has extensively studied and developed data integration approaches and systems leading to, among others, a virtual approach to data
integration called mediation [40]. Several systems and prototypes have been developed: examples of such systems are TSIMMIS [16], PICSEL [18], Information Manifold [23], AGORA [26], Styx [1]. A mediation system provides to the user a uniform
interface of the different data sources via a common data model. Schema integration
issues are the sources heterogeneity, global schema modeling and definition, the definition and the management of the mapping rules that express the correspondences
between the global schema and the data source ones, the source semantics and the
schema evolution.
There are two main approaches to data mediation: in the global as view (GAV)
approach, the global schema is defined as a set of views over local schemas, while
in the local as view (LAV) one, local sources are defined as a set of views over a
given global schema, pertaining sometimes to some domain ontology. Pros and cons
of these two approaches are well-known: query rewriting is straightforward in GAV
while adding a new data source is made easy in LAV. Alternative approaches, such
as global local as view (GLAV) [14] or both as view (BAV) [6], have been proposed. GLAV combines the expressive power of both LAV and GAV, allowing flexible schema definitions; BAV is based on the use of reversible schema transformation
sequences.
The fundamental question, when attempting to interoperate several data sources,
is twofold: on the one end, the identification of different objects having a semantic link, and coming from different data sources, on the other end, the resolution of structural (schematic) differences between objects having a semantic link.
Schematic conflicts between data may arise when equivalent concepts are represented differently in local data sources. Those conflicts can be associated with concept names, their data types (e.g. a building can be a polygon in a data source
and a point in another), the unit (perimeter can be expressed in meters in a data
source and in kilometers in another one), the attributes (some attributes may be absent in some data sources). Another kind of schematic conflict is the difference in
the representation of an attribute. For example, one can represent an address with
a single attribute and another can represent it with a tuple (Number, Street, City,
Zip code).
In addition to the query language, the power of an integration system is based
on how schema mapping is performed and how efficient is the query rewriting algorithm. Most of the existing mediation systems use views to express correspondences
between the real data source schemas and the integrated one. For example, in Information Manifold, the real data sources are expressed as a set of relational views over
Omar Boucelma, Mehdi Essid, and Yassine Lassoued
7.2 Related Work
In this section, we describe some related work that will help the reader to understand our approaches to both spatial data integration and geographical data quality
management.
7.2.1 Data Integration
The DataBase (DB) community has extensively studied and developed data integration approaches and systems leading to, among others, a virtual approach to data
integration called mediation [40]. Several systems and prototypes have been developed: examples of such systems are TSIMMIS [16], PICSEL [18], Information Manifold [23], AGORA [26], Styx [1]. A mediation system provides to the user a uniform
interface of the different data sources via a common data model. Schema integration
issues are the sources heterogeneity, global schema modeling and definition, the definition and the management of the mapping rules that express the correspondences
between the global schema and the data source ones, the source semantics and the
schema evolution.
There are two main approaches to data mediation: in the global as view (GAV)
approach, the global schema is defined as a set of views over local schemas, while
in the local as view (LAV) one, local sources are defined as a set of views over a
given global schema, pertaining sometimes to some domain ontology. Pros and cons
of these two approaches are well-known: query rewriting is straightforward in GAV
while adding a new data source is made easy in LAV. Alternative approaches, such
as global local as view (GLAV) [14] or both as view (BAV) [6], have been proposed. GLAV combines the expressive power of both LAV and GAV, allowing flexible schema definitions; BAV is based on the use of reversible schema transformation
sequences.
The fundamental question, when attempting to interoperate several data sources,
is twofold: on the one end, the identification of different objects having a semantic link, and coming from different data sources, on the other end, the resolution of structural (schematic) differences between objects having a semantic link.
Schematic conflicts between data may arise when equivalent concepts are represented differently in local data sources. Those conflicts can be associated with concept names, their data types (e.g. a building can be a polygon in a data source
and a point in another), the unit (perimeter can be expressed in meters in a data
source and in kilometers in another one), the attributes (some attributes may be absent in some data sources). Another kind of schematic conflict is the difference in
the representation of an attribute. For example, one can represent an address with
a single attribute and another can represent it with a tuple (Number, Street, City,
Zip code).
In addition to the query language, the power of an integration system is based
on how schema mapping is performed and how efficient is the query rewriting algorithm. Most of the existing mediation systems use views to express correspondences
between the real data source schemas and the integrated one. For example, in Information Manifold, the real data sources are expressed as a set of relational views over