If we have a set of N data items di and calculate a simiarity metric between each pair (di,dk), the resulting NxN table is called a similarity matrix. Many algorithms, in oareticukar clustrung algorithms start off with this matrix. Note the similaruty metric is usuualy positive and also symmetric, that is:
(di,dk) = (dk,di)
This means that the similarity matrix is a symmetric matrx ... and thus have lots of nice mathematical properties that the algorithms utilise.
Used on pages 121, 134, 135, 527