1
$\begingroup$

I have performed LDA using the topicmodels package in R. When I was inspecting my theta values for each document in the corpus (~450,000 documents), I noticed that the average max theta, or in other words, the average highest probability that the document is in a certain topic, was ~9%. This seems very low, but I am unable to find any literature that would give me an idea of what it should be around.

In case I am doing something else wrong, here are the parameters I am using:

burnin <- 1000
iter <- 1000
keep <- 10
k<-100 
models <- LDA(dtm2, k, method = "Gibbs", control = list(burnin = burnin,        
iter = iter, keep = keep))

phi <- posterior(models)$terms %>% as.matrix
theta <- posterior(models)$topics %>% as.matrix
$\endgroup$

1 Answer 1

1
$\begingroup$

You should decrease the value of alpha. By default, the alpha value is 50/k where k is the number of topics.

When you decrease the alpha value, the probability of a document being represented by fewer topics increases.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.