Similarity Metric

• One simple similarity metric is Euclidean Distance.
• Remember the Pythagorean Theorem.
• A^2+B^2 = C^2
• This works in N dimensions.
• In the bank example, the distance between the three tuples that describe the cases is a simple similarity metric.
• Is it generalizable?
• You could also use a city-block metric.
• Similarity and Search Spaces
• The key is that the similarity metric needs to work.
• That is, in the similarity space, cases with the same answer need to cluster together.
• With Euclidean or city-block distance, you can make features more important by multiplying their values by a constant greater than one for the similarity metric.
• With scalars, you can make a table to say how similar features are.
• You can also take advantage of ontologies and semantic distance in a semantic net.