Similarity Queries in Data Bases Using Metric Distances - from Modeling Semantics to Its Maintenance

Similarity queries in traditional databases work directly on attribute values. But, often similar attribute values do not indicate similar meanings. Semantic background information is needed to enhance similarity query performance. In this paper a method will be addressed which follows the idea to map attribute values to multidimensional points and then interpret the distances between that points as similarity. The second part brings the questions “How to arrange these points that they correspond to real world?” and “Can that be done automatically?” into focus and comes to the following result: For the case that all similarities are known in advance a good solution is given otherwise it turns to a complex optimization problem.