Retrieve

Retrieving a case means to start with a (partial) new case, and retrieve the best matching previous case. It involves the following subtasks:

• Identify features - this may simply be to notice the feature values for a case, or can be a more complex evaluation which tries to ``understand'' the problem in a context. This can be filtering out noisy problem descriptors, infer other relevant problem features, check whether the feature values make sense in the given context, or generate expectations of other feature values. The checking might be done against a knowledge model (cases or general knowledge) or by asking the user.
• Initially match - usually done in two parts, first an initial matching process which gives a list of possible candidates, which are then further examined to select the best. There are three ways of retrieving a case or a set of cases: By following direct index pointers from the problem features, by searching an index structure, or by searching in a model of domain knowledge. Matching cases can be found by comparing with input features or input features and features inferred from others using domain knowledge. The features can be compared using some similarity measure, which is usually normalized, for instance to the range [0,1], so that it is easy to compare cases based on several or all features. The Case-Based reasoner can try to ``understand'' the problem, and use this understanding in comparing. It can also weigh the input features.
• Select - select a best match from the cases returned by the initially match. The reasoner tries to explain away non-identical features. If the match is not good enough, a better one is sought by using links to closely related cases. The selection process can generate consequences and expectations from each retrieved case, by using an internal model or by asking the user.