How to Measure Topic Coherence

by Haridas Narayanaswamy
Unsupervised topic modeling algorithms like LDA and NMF produces list of vocabularies for each topic after the training. These vocabs help human to assign the subject information of the topic model. So how we measure the quality of these topic words ?, this problem has to be addressed in unsupervised topic clustering algorithms like LDA / NMF to understand models are improving or not. It’s always a challenge to qualitatively measure the goodness of the words produced by each topic.