plotTokens: Plot probable tokens for a given topic

In kshirley/LDAtools: Tools to fit a topic model using Latent Dirichlet Allocation (LDA)

Description

Plot probable tokens for a given topic

Usage

plotTokens(phi = vector(), vocab = character(), n.tokens = 20,
  lambda = 0.5, p = vector(), ...)

Arguments

phi	numeric vector with the probability of each token for a given topic.
vocab	character vector of the vocabulary for the corpus
n.token	the number of tokens to plot, where the default is n.token = 20.
lambda	a parameter between 0 and 1 to control how tokens are ranked within topics
p	the marginal probabilities of the tokens in the vocabulary
...	additional arguments to the plot() function

Details

The ranking of tokens within topics is based on a weighted average of the probability of a token (given the topic) and the lift, where the lift of a token is defined as the probability of the token (given the topic) divided by the marginal probability of the token (i.e. across all topics). The ranking that determines the top n.token tokens to plot is simply lambda * log(p(token)) + (1 - lambda) * log(p(token | topic)/p(token)).

Note: the ordering of phi, vocab, and p must be the same (i.e. the nth element of each vector must correspond to the same token)

Examples

data(APinput)
data(APtopics) #load output instead for demonstration
probs <- getProbs(word.id=APinput$word.id, doc.id=APinput$doc.id, topic.id=APtopics$topics,
               vocab=APinput$vocab)
 #THE ORDERING OF phi, vocab and p MUST MATCH!
tokens <- factor(APinput$vocab[APinput$word.id], levels=colnames(probs$phi.hat))
token.tab <- table(tokens)
p <- token.tab/sum(token.tab)
plotTokens(phi=probs$phi.hat[1,], vocab=names(p), n.tokens=30, lambda=1/3, p)
# plot all the topics!
## Not run: 
 for (i in seq_along(probs$phi.hat[,1])) {
   plotWords(probs$phi.hat[i,], tokens)
 }

## End(Not run)

kshirley/LDAtools documentation built on May 20, 2019, 7:03 p.m.