Skip to contents

Sum over the top-M words of each topic of log((D(w_i,w_j)+1)/D(w_j)), using document co-occurrence counts. Higher (less negative) is more coherent.

Usage

semantic_coherence(model, M = 10L)

Arguments

model

A faSTM fit (must carry its document-term matrix; faSTM stores it).

M

Number of top words per topic.

Value

A numeric vector, one coherence value per topic.