Add topic correlation based on document distribution

[clovis]

We currently compute correlation based on frequency over time. We should also compute similarity based on document distribution by transposing the document-topic matrix.