Databases Reference
In-Depth Information
attribute correspondence selection process. Such applications involve data exchange
in data spaces and integration in the Semantic Web. Matching requires more automa-
tion, as the time frame for reaching matching decisions shrinks. Also, the role of
the human designer is changing, as data is being scraped from sources to which
no human designer is available for evaluating the matching correctness. Finally,
the sheer amount of data integration decisions that need to be taken require the
development of methods for nonexperts.
As a result of this setting change, several research attempts were devoted to
improve attribute correspondences by extending their capabilities, as illustrated in
Fig.
3.1
. In this chapter, we describe recent advances in the schema matching lit-
erature that attempt to enhance the capabilities of attribute correspondences. We
discuss contextual schema matching (Sect.
3
) that can express that attribute corre-
spondences hold under certain instance conditions. For example, a
code
attribute in
one schema may refer to an
ISBN
attribute in another schema only if the sold item is
a topic. Semantic matching is introduced in Sect.
4
, extending attribute correspon-
dences to support ontological constructs. Traditionally, attribute correspondences
are limited to the ontological construct known as
synonym
. However, correspon-
dences may be of various types. For example, a pair of attributes may correspond
through a
hyperonym
relationship, where one attribute represents a more general
concept than another. As a concrete example, consider two databases for selling
electronic products. The correspondence between an attribute
digitalCamera
and
an attribute
PhotoAndCamera
is that of subsumption and not equivalence. Finally,
probabilistic schema matching (Sect.
5
) extends attribute correspondences to con-
currently assume the validity of attribute correspondences, which were considered
conflicting in the past. For example, a 1
W
1 matching constraint may hold valid, yet
under the probabilistic approach an attribute may (probabilistically) be associated
Attribute
A
Attribute
B
Contextual
Attribute
Correspondence
Probabilistic
Attribute
Correspondence
Semantic
Attribute
Correspondence
Attribute
B1
0.3
R.Credit Cardlnfo.type
= 'RoomsRUs'
Attribute
B
Attribute
A
0.5
Attribute
B2
Attribute
A
0.2
hyperonym
Attribute
A
Attribute
B
Attribute
B3
Fig. 3.1
Extensions to attribute correspondences
