A study of the similarities between topics (Q2488418)

Let \(A_i\), \(i=1,\dots,I\), be random events. The authors propose to use \[ k(A_i,A_j)=P(A_i\cap A_j)-P(A_i)P(A_j) \] as a measure of ``probabilistic similarity'' of \(A_i\) and \(A_j\). As a graphical tool for investigation of similarities, a plot of points \(\{(P(A_i),k(A_j,A_i)\}\) is proposed. This technique is applied to data on the frequencies of twelve topics related to machine learning (such as ``neural network learning'' or ``kernel methods'') in Web pages from the Google database.