Question

Different data types may need to use different similarity (distance) measures. What is the expected similarity measure in each of the following applications:

(a) clustering stars in the universe,

(b) clustering text documents,

(c) clustering clinical test data, and

(d) clustering houses to find delivery centers in a city with rivers and bridges?

Answer 1

(a) Euclidean distance (spatial and interval based) - straight-line distance between two points in Euclidean space

(b) cosine (vector data) similarity - similarity between two non-zero vectors of an inner product space that measures cosine of angle between them.

(c) asymmetric data (example- Jaccard coefficient) - comparing the similarity and diversity of sample sets

(d) reachable distance (not use direct Euclidean distance), consider the obstacles- situated within easy reach