Please note that this page is still work-in-progress.

Side information for ProCAKE mechanisms #

This page contains the following content:

Similarity #

Similarity and Utility #

Many retrieval methods follow a simple text matching approach without using knowledge about the semantics behind. The consequences may be fatal. For example, if an available product matches the requirements of the requested product but is not proposed by the retrieval service, the opportunity for improving efficiency, quality, or time-to-market is lost. On the other hand, if a proposed product is not sufficient resources are wasted. CBR is able to minimise both potential failure classes, but it strongly depends on the quality of the similarity computation.

The goal of the similarity measures is to approximate the utility. Utility describes how helpful a solution is for a new problem. Thereby the traditional binary perception (true / false) is replaced by a finer grading. The approximation through the similarity measures bases on the assumption that similar problems have similar solutions or in other words, the solution of a problem is also useful for similar problems.

The problem descriptions of previous experiences get successively compared to the new problem by using the similarity function. The resulting similarity values are used to rank the cases in the case-base and the most similar cases are used to adapt their solutions to a new solution for the new problem. Consequently, a similarity measure sim is a function that expresses the similarity between two problem descriptions as a numerical value. The higher the value, the higher is the similarity between the two cases.

Depending on the representation format several similarity measures can be defined. For the attribute value representation the local-global principle is a well established technique. Thereby, the similarity calculation is performed by combining similarity values for each single attribute of the attribute space.

The local similarity values are used by the global similarity measure to aggregate them into one similarity value for all attributes. Thereby, the importance, relevance, and utility aspect of each attribute is encoded into the function.