LDAvis

LDAvis

Interactive visualization of LDA topic models.

Basic Usage

library(LDAvis)
library(topicmodels)

# Fit LDA model first
lda_model <- LDA(dtm, k = 10)

# Create visualization
json <- createJSON(
  phi = posterior(lda_model)$terms,
  theta = posterior(lda_model)$topics,
  doc.length = rowSums(as.matrix(dtm)),
  vocab = colnames(dtm),
  term.frequency = colSums(as.matrix(dtm))
)

# View in browser
serVis(json)

With topicmodels

library(topicmodels)
library(LDAvis)

# Fit LDA
lda <- LDA(dtm, k = 10, method = "Gibbs",
           control = list(seed = 123, iter = 1000))

# Extract components
phi <- posterior(lda)$terms
theta <- posterior(lda)$topics
vocab <- colnames(dtm)
doc_length <- rowSums(as.matrix(dtm))
term_freq <- colSums(as.matrix(dtm))

# Create JSON
json <- createJSON(
  phi = phi,
  theta = theta,
  doc.length = doc_length,
  vocab = vocab,
  term.frequency = term_freq
)

serVis(json)

Customization

json <- createJSON(
  phi = phi,
  theta = theta,
  doc.length = doc_length,
  vocab = vocab,
  term.frequency = term_freq,
  R = 30,                    # Number of terms to display
  lambda.step = 0.01,        # Lambda slider step
  mds.method = jsPCA,        # MDS method
  cluster = NULL,            # Cluster topics
  reorder.topics = TRUE      # Reorder by prevalence
)

Save Visualization

# Save as HTML
serVis(json, out.dir = "lda_vis", open.browser = FALSE)

# Creates:
# - lda_vis/index.html
# - lda_vis/lda.json
# - lda_vis/d3.v3.js
# - lda_vis/ldavis.v1.0.0.js
# - lda_vis/ldavis.v1.0.0.css

With text2vec

library(text2vec)
library(LDAvis)

# Create DTM with text2vec
it <- itoken(texts, preprocessor = tolower, tokenizer = word_tokenizer)
vocab <- create_vocabulary(it)
vectorizer <- vocab_vectorizer(vocab)
dtm <- create_dtm(it, vectorizer)

# Fit LDA
lda_model <- LDA$new(n_topics = 10)
doc_topic_distr <- lda_model$fit_transform(dtm, n_iter = 1000)

# Get components
phi <- lda_model$get_top_words(n = ncol(dtm), lambda = 1)
# ... create visualization