110
Matt Duckham and Mike Worboys
• indicate initial techniques to adapt the automated fusion process to operate in the
presence of imperfect and uncertain geographical information.
Following a brief motivational example (Sect. 6.1.1), Sect. 6.2 presents a review
of the relevant literature. Section 6.3 then sets out the foundations of inductive inference for automated geographical information fusion. The limitations resulting from
the unreliability of the inductive reasoning process are set out in Sect. 6.4, while
Sect. 6.5 addresses the management of uncertainty in the input geographical data.
Finally, Sect. 6.6 concludes the chapter with a look at future research.
6.1.1 Motivational Example
Data on the structural characteristics of buildings is often important to decision
makers as part of an emergency response effort. It is not unusual for several different agencies to collect such data for the same geographical region and to make
it available online. These agencies may use heterogeneous definitions or may even
produce semistructured data without separate or fixed definitions (see Chap. 2). For
example, the category “Reinforced concrete building” in spatial database A may not
have the same meaning as the category “Non-wooden building” in spatial database
B (this example is taken from a study of the 1995 Kobe Earthquake [64]). Current
geographical information fusion techniques rely on the generation of a manual specification of the semantic relationships between different categories by a human domain expert. Such manual techniques can be slow, unreliable, and do not scale easily
to Web-based information fusion scenarios.
However, if all the instances of buildings categorized as “Reinforced concrete
building” in spatial database A are categorized as “Non-wooden building” in spatial
database B, then this provides evidence that the category “Reinforced concrete building” is a subcategory of “Non-wooden building.” Although this example is highly
simplified, it does support the central intuition behind using inductive inference for
geographical information fusion: that analysis of spatial relationships can be used to
infer semantic relationships. It is important to note that this inference process does
not necessarily require an understanding of the meaning of “Non-wooden building”
or “Reinforced concrete building,” and hence can be applied in the context of automated reasoning systems.
6.2 Background
The semantics of an information source may be described using an ontology (defined
as “an explicit specification of a conceptualization” [32]). The task of fusing information compiled using different ontologies is a classical problem in information science
(e.g. [69]), and continues to be a highly active research issue within many topics,
including databases [40, 42, 57], interoperability [56, 65], the semantic Web [6, 18],
medical information systems [28, 55], knowledge representation [10], data warehousing [68, 73], and, of course, geographical information fusion (Sect. 6.2.3).
Matt Duckham and Mike Worboys
• indicate initial techniques to adapt the automated fusion process to operate in the
presence of imperfect and uncertain geographical information.
Following a brief motivational example (Sect. 6.1.1), Sect. 6.2 presents a review
of the relevant literature. Section 6.3 then sets out the foundations of inductive inference for automated geographical information fusion. The limitations resulting from
the unreliability of the inductive reasoning process are set out in Sect. 6.4, while
Sect. 6.5 addresses the management of uncertainty in the input geographical data.
Finally, Sect. 6.6 concludes the chapter with a look at future research.
6.1.1 Motivational Example
Data on the structural characteristics of buildings is often important to decision
makers as part of an emergency response effort. It is not unusual for several different agencies to collect such data for the same geographical region and to make
it available online. These agencies may use heterogeneous definitions or may even
produce semistructured data without separate or fixed definitions (see Chap. 2). For
example, the category “Reinforced concrete building” in spatial database A may not
have the same meaning as the category “Non-wooden building” in spatial database
B (this example is taken from a study of the 1995 Kobe Earthquake [64]). Current
geographical information fusion techniques rely on the generation of a manual specification of the semantic relationships between different categories by a human domain expert. Such manual techniques can be slow, unreliable, and do not scale easily
to Web-based information fusion scenarios.
However, if all the instances of buildings categorized as “Reinforced concrete
building” in spatial database A are categorized as “Non-wooden building” in spatial
database B, then this provides evidence that the category “Reinforced concrete building” is a subcategory of “Non-wooden building.” Although this example is highly
simplified, it does support the central intuition behind using inductive inference for
geographical information fusion: that analysis of spatial relationships can be used to
infer semantic relationships. It is important to note that this inference process does
not necessarily require an understanding of the meaning of “Non-wooden building”
or “Reinforced concrete building,” and hence can be applied in the context of automated reasoning systems.
6.2 Background
The semantics of an information source may be described using an ontology (defined
as “an explicit specification of a conceptualization” [32]). The task of fusing information compiled using different ontologies is a classical problem in information science
(e.g. [69]), and continues to be a highly active research issue within many topics,
including databases [40, 42, 57], interoperability [56, 65], the semantic Web [6, 18],
medical information systems [28, 55], knowledge representation [10], data warehousing [68, 73], and, of course, geographical information fusion (Sect. 6.2.3).