Generation of similarity measures from different sources

Abstract

Knowledge that quantifies the similarity between complex objects forms a vital part of problem-solving expertise within several knowledge intensive tasks. This paper shows how implicit knowledge about object similarities is made explicit in the form of a similarity measure. The development of a similarity measure is highly domain-dependent. We will use the domain of fluidic engineering as a complex and realistic platform to present our ideas. The evaluation of the similarity between two fluidic circuits is needed for several tasks: (i) Design problems can be supported by retrieving an existing circuit which resembles an (incomplete) circuit description. (ii) The problem of visualizing technical documents can be reduced to the problem of arranging similar documents with respect to their similarity. The paper in hand presents new approaches for the construction of a similarity function: Based on knowledge sources that allow for an expert-friendly knowledge acquisition, machine learning is used to compute an explicit similarity function from the acquainted knowledge.