Get some topic labels using a "more probable" method of terms

Source: R/topic_modeling_utilities.R

Function calls GetProbableTerms with some rules to get topic labels. This function is in "super-ultra-mega alpha"; use at your own risk/discretion.

LabelTopics(assignments, dtm, M = 2)

Arguments

assignments

A documents by topics matrix similar to theta. This will work best if this matrix is sparse, with only a few non-zero topics per document.

dtm

A document term matrix of class matrix or dgCMatrix. The columns of dtm should be n-grams whose colnames have a "_" where spaces would be between the words.

M

The number of n-gram labels you want to return. Defaults to 2

Value

Returns a matrix whose rows correspond to topics and whose j-th column corresponds to the j-th "best" label assignment.

Examples

# make a dtm with unigrams and bigrams
data(nih_sample_topic_model)

m <- nih_sample_topic_model

assignments <- t(apply(m$theta, 1, function(x){
  x[ x < 0.05 ] <- 0
  x / sum(x)
}))

assignments[is.na(assignments)] <- 0

labels <- LabelTopics(assignments = assignments, dtm = m$data, M = 2)
#> dtm does not appear to contain ngrams. Using unigrams but ngrams will work much better.