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 suppported by retrievimg 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 knowl- edge acquisition, machine learning is used to compute an explicit similarity func- tion from the acquainted knowledge.