demo/lda.R

In lda: Collapsed Gibbs Sampling Methods for Topic Models

require("ggplot2")
require("reshape2")
data(cora.documents)
data(cora.vocab)

theme_set(theme_bw())
set.seed(8675309)

K <- 10 ## Num clusters
result <- lda.collapsed.gibbs.sampler(cora.documents,
                                      K,  ## Num clusters
                                      cora.vocab,
                                      25,  ## Num iterations
                                      0.1,
                                      0.1,
                                      compute.log.likelihood=TRUE) 

## Get the top words in the cluster
top.words <- top.topic.words(result$topics, 5, by.score=TRUE)

## Number of documents to display
N <- 10

topic.proportions <- t(result$document_sums) / colSums(result$document_sums)
topic.proportions <-
  topic.proportions[sample(1:dim(topic.proportions)[1], N),]
topic.proportions[is.na(topic.proportions)] <-  1 / K

colnames(topic.proportions) <- apply(top.words, 2, paste, collapse=" ")

topic.proportions.df <- melt(cbind(data.frame(topic.proportions),
                                   document=factor(1:N)),
                             variable.name="topic",
                             id.vars = "document")  

qplot(topic, value, fill=document, ylab="proportion",
      data=topic.proportions.df, geom="bar", stat="identity") +
  theme(axis.text.x = element_text(angle=90, hjust=1)) +
  coord_flip() +
  facet_wrap(~ document, ncol=5)