mallet (version 1.3.0)

MalletLDA: Create a Mallet topic model trainer

Description

This function creates a java cc.mallet.topics.RTopicModel object that wraps a Mallet topic model trainer java object, cc.mallet.topics.ParallelTopicModel. Note that you can call any of the methods of this java object as properties. In the example below, I make a call directly to the topic.model$setAlphaOptimization(20, 50) java method, which passes this update to the model itself.

Usage

MalletLDA(num.topics = 10, alpha.sum = 5, beta = 0.01)

Value

a cc.mallet.topics.RTopicModel object

Arguments

num.topics

The number of topics to use. If not specified, this defaults to 10.

alpha.sum

This is the magnitude of the Dirichlet prior over the topic distribution of a document. The default value is 5.0. With 10 topics, this setting leads to a Dirichlet with parameter \(\alpha_k = 0.5\). You can intuitively think of this parameter as a number of "pseudo-words", divided evenly between all topics, that are present in every document no matter how the other words are allocated to topics. This is an initial value, which may be changed during training if hyperparameter optimization is active.

beta

This is the per-word weight of the Dirichlet prior over topic-word distributions. The magnitude of the distribution (the sum over all words of this parameter) is determined by the number of words in the vocabulary. Again, this value may change due to hyperparameter optimization.

Examples

if (FALSE) {
# Read in sotu example data
data(sotu)
sotu.instances <-
   mallet.import(id.array = row.names(sotu),
                 text.array = sotu[["text"]],
                 stoplist = mallet_stoplist_file_path("en"),
                 token.regexp = "\\p{L}[\\p{L}\\p{P}]+\\p{L}")

# Create topic model
topic.model <- MalletLDA(num.topics=10, alpha.sum = 1, beta = 0.1)
topic.model$loadDocuments(sotu.instances)

# Train topic model
topic.model$train(200)

# Extract results
doc_topics <- mallet.doc.topics(topic.model, smoothed=TRUE, normalized=TRUE)
topic_words <- mallet.topic.words(topic.model, smoothed=TRUE, normalized=TRUE)
top_words <- mallet.top.words(topic.model, word.weights = topic_words[2,], num.top.words = 5)
}