Orders topics according to similarity in topic distributions
In this procedure, for each topic, the probability distribution over documents is calculated. For each topic pair (i,j), the (symmetrized) KL distance between topic distributions i and j is calculated. The KL distances are then subjected multidimensional scaling (MDS). The first dimension of the MDS solution is then used to get an ordering for topics.
Notes: The variable ALPHA is the hyperparameter on the topic distributions needed to calculate the topic-document distributions calculate prob distribution over documents for each topic