Database Reference
In-Depth Information
•
Translation of the
constraints on the data
by inheritance, a rule is generated to express the
fact that their references cannot be reconciled. A
transitivity rule allows inferring new reconcilia-
tion decisions by applying transitivity on the set
of already inferred reconciliations.
See (Saïs et al., 2009) for a complete descrip-
tion of the generation process of reconciliation
rules.
sources
The UNA assumption, if it is stated on the
sources
S
1
and
S
2
, is translated automatically by
four rules. For example, the following rule R1 ex-
presses the fact that two distinct references coming
from the same source cannot be reconciled.R1:
Src1(x) ∧ Src1(y) ∧ (x≠y) ⇒ ¬ Reconcile(x,y)
where
Src
i
(x)
means that the reference
x
is coming
from a source
S
i
.
Analogous rules express that one reference
coming from a source
S
i
can be reconciled with
at most one reference coming from a source
S
j
.
Similarly, two rules are generated for translating
LUNA semantics.
Reasoning Method for Reference Reconcili-
ation
In order to infer sure reconciliation and non-
reconciliation decisions, we apply an automatic
reasoning method based on the resolution prin-
ciple (Robinson, 1965; Chang & Lee, 1997). This
method applies to the clausal form of the set of rules
R
described above and a set of facts
F
describing
the data, which is generated as follows.
•
Translation of the schema
constraints.
For each relation
R
declared as functional by
the constraint PF(
R
), the following rule R6.1(
R
)
is generated:R6.1(R): Reconcile(x, y) ∧ R(x, z)
∧ R(y, w) ⇒ Reconcile(z, w)
For example, the following rule is generated
concerning the relation
located
which relates refer-
ences of cultural places to references of addresses
and which is declared functional:R6.1(located):
Reconcile(x, y) ∧ located(x, z) ∧ located(y, w)
⇒ Reconcile(z, w)
For each attribute
A
declared as functional by
the axiom PF(
A
), a similar rule which concludes
on
SynVals
is generated.
Likewise, analogous rules are generated for each
relation
R
and each attribute
A
declared as inverse
functional. Rules are also generated for translating
combined constraints PF(
P
1
,..., P
n
) and PFI(
P
1
,...,
P
n
) of (inverse) functionality. For example, the
declaration PFI(
paintedBy, paintingName
) states a
composed functional dependency which expresses
that the artist who painted it jointly with its name
functionally determines a painting.
For each pair of classes
C
and
D
involved
in a DISJOINT(
C,D
) statement declared in the
schema, or such that their disjunction is inferred
•
Generation of the set of facts.
The set of RDF facts corresponding to the
description of the data in the two sources
S1
and
S2
is augmented with the generation of:
•
new class-facts, relation-facts and attri-
bute-facts derived from the domain and
range constraints that are declared in RDFS
for properties, and from the subsumption
statements ;
facts of the form
•
Src
1
(i)
and
Src
2
(j)
;
synonymy facts of the form
•
SynVals(v
1
,v
2
)
for each pair
(v
1
,v
2
)
of basic values that are
identical (up to some punctuation or case
variations) ;
non
•
synonymy
facts
of
the
form
¬SynVals(v
1
,v
2
)
for each pair
(v
1
,v
2
)
of
distinct basic values of a functional attri-
bute for which it is known that each pos-
sible value has a single form. For instance,
¬SynVals(“France”, “Algeria”)
can be
added.
Resolution-based algorithm for reference
•
reconciliation.
Search WWH ::
Custom Search