Database Reference
In-Depth Information
Figure 3. Illustration of the effect of the grip factor g in an Euclidean 2-dimensional space, considering
|Q| = 3
of the maximum distance. Figure 3 helps on getting the intuition of the meaning of these grip factors
presenting the effect of
g
in a 2-dimensional Euclidean space and considering
Q
composed of the three
query centers shown. Each curve represents the geometric place where Equation 1 has the same value.
Therefore, each curve is an isoline representing a different covering radius, thus defining both range
and
k
-limited queries.
There are several applications that can benefit from the use of queries that employ this type of
similarity predicates. For instance, multimedia data (such as images, audio or video) require extracting
features that are used in place of the data element when performing the comparisons in a similarity
query. The features are usually the result of mathematical algorithms, resulting in low level features.
Considering an image domain, the features are usually based on color, texture and shape. However, there
exists a semantic gap between the low level features and the human interpretation subjectivity. To deal
with the semantic gap, relevance feedback techniques have been developed. In these techniques, positive
and/or negative examples are informed by the user to allow the system to derive a more precise repre-
sentation of the user intent (Zhou and Huang, 2003). The new representation of the user intent can be
achieved by query point movement or by multiple point movement techniques. In these techniques, the
system learns from the search results provided by the user and takes advantage of this information to
adapt ranking functions. One way to tell to the system what is the user's intention is specifying, in the
same query, other elements besides the query center, which are positive or negative examples of the
intend answer. This representation is based on multiple query centers.
Similarity Joins
Binary operators correspond to similarity joins. A similarity join over an attribute
S
j
in relation
R
1
and
<
join
−
predicate
>
an attribute
S
k
in relation
R
2
can be represented as (
R
⊗
R
), where the join predicate is of the
1
2
form
S
q and
θ
denotes a similarity operator. This predicate expresses a similarity comparison
between the set of values
S
j
S
j
k
Ì of an attribute
S
j
from the relation
R
1
and a set of values
S
k
Ì of
j
k
