Databases Reference
In-Depth Information
Fig. 10.5
AgreementMaker user interface for testing and comparing combination strategies
AgreementMaker provides a combination strategy based on the linear interpola-
tion of the similarity values. The weights can be either user assigned or evaluated
through automatically determined quality measures. The system allows for serial
and parallel composition where, respectively, the output of one or more methods
can be used as input to another one, or several methods can be used on the same
input and then combined.
5.3.2
Harmony
Harmony schema matcher [
Mork et al. 2008
;
Smith et al. 2009
] combines multi-
ple matcher algorithms by using a vote merger. The vote merging principle is a
weighted average of the match scores provided by each match voter. A match voter
provides a confidence score for each pair of schema elements to be matched. Then,
Similarity Flooding strategy [
Melnik et al. 2002
] is applied to adjust the confidence
scores based on structural information. Thus, positive confidence scores propagate
up the graph. An interesting feature of Harmony lies in its graphical user interface
for viewing and modifiying the discovered schema correspondences. This allows
to assist the users to focus their attention on different ways. This assistance is
done through a filter, which is a predicate that is evaluated against each candidate
correspondence. Harmony supports two kinds of filters. The first kind named
link
filters
depends on the characteristics of a candidate correspondence. For example,
applying the confidence filter will have an effect to graphically display those corre-
spondences, whose match score falls within the specific range of values. The second
one named
node filters
is related to a schema element characteristics. This kind of
