Information Technology Reference
In-Depth Information
the frequently-used methods for case retrieval are: nearest neighbor, induction,
and template retrieval.
(1) Nearest neighbor: This approach assesses the similarity between stored
cases and the new input case based on matching a weighted sum of
features. The key point here is to determine the weights of the features.
One limitation of this approach is the time complexity increases with the
case number of case base linearly. Therefore this approach is more
effective when the case base is relatively small.
(2) Induction: Induction algorithms determine which features do the best job in
discriminating cases, and generate a decision tree to organize the cases in
memory effectively. This approach is useful when a single case feature is
required as a solution, and where that case feature is dependent upon
others.
(3) Template retrieval: It is similar to SQL-like queries, and returns all cases
that fit within certain parameters. This technique searches all examples that
can return within the range of certain parameter value and is often used
before other techniques, such as nearest neighbor, to limit the search space to
a relevant section of the case base.
5.7 Similarity Relations in CBR
The process of retrieving the most similar cases to the current problem or
situation is especially critical in that the successful retrieval of cases with
high quality is the prerequisite of successful applications. Due to the fact that
case retrieval is carried on the basis of similarity, the successful retrieval of
similar cases is totally determined by the definition of “similarity”. We can
not get useful cases using an inappropriate similarity measure assessment
among the cases, let alone successful applications. Similarity is a core concept
in case-based reasoning.
As indicated by case representation, a case comprises many attributes, and
the similar degrees among the cases are defined in terms of the similarity
degrees among the component attributes. Among well known similarities are:
semantic similarity, structural similarity, goal similarity, and individual
similarity.
