Topic modelling

The function to create and train and an LDA model.

Usage

topicsModel(
  dtm,
  num_topics = 20,
  num_top_words = 10,
  num_iterations = 1000,
  seed = 42
)

Arguments

dtm

(R_obj) The document term matrix -> output of topicsDtm function

num_topics

(integer) The number of topics to be created

num_top_words

(integer) The number of top words to be displayed

num_iterations

(integer) The number of iterations to run the model

seed

(integer) A seed to set for reproducibility

Value

A named list containing the following elements:

name

Description

instances

Java object reference: A list of all documents used for topic modeling, in which each document is preprocessed (e.g., tokenized and vectorized). This object is part of the Mallet package's internal structure.

inferencer

Java object reference: This is the topic inferencer, which allows the inference of topic distributions for new, unseen documents based on the trained model.

top_terms_mallet

A data frame containing the top terms of each topic, showing which concepts each topic likely represents. The number of top terms shown here can be adjusted with the argument num_top_words.

top_terms

A data frame containing the top terms of each topic, showing which concepts each topic likely represents. The number of top terms shown here can be adjusted with the argument num_top_words.

phi

A matrix of the topic-word distribution: Each row represents a topic, and each column represents a word from the document term matrix. The values show the probability of a word given a topic P(word|topic).

topic_docs

A matrix of document-topic distribution: Each row represents a document, and each column represents a topic. The values show the probability of a topic given a document, P(topic|document).

frequencies

A data frame of term frequencies. word = every word in the document term matrix, word.freq = the frequency of each word across all documents, doc.freq = the number of documents in which each word appears.

vocabulary

A character vector of all unique terms in the document term matrix.

labels

A list of topic labels. These short labels are the most representative term for each topic, making it easy to identify and understand them.

theta

A data frame of document-topic probabilities: each row represents a document, and each column represents a topic. Similar to topic_docs, this shows the contribution of each topic to each document. Each row sums to 1, representing the document’s composition of topics.

prevalence

A numeric vector showing the overall prevalence (prominence) of each topic in the corpus. The prevalences are expressed as percentages relative to the other topics and add up to 100 Higher values indicate topics that are present in more documents.

coherence

A numeric vector showing the coherence of each topic. Coherence scores indicate how semantically consistent and interpretable the topics are. Higher coherence generally indicates better-quality topics.

pred_model

A list containing components of the predictive model, including phi (word-topic probability matrix), theta (document-topic probabilities matrix), alpha (Dirichlet prior of topics), gamma (hyperparameters of word-topic assignments), and data (sparse matrix representing the document term matrix.)

dtm_settings

A list of settings used for preprocessing and building the document term matrix (dtm), including n-gram ranges, stopword removal, frequency thresholds, and random seed settings.

summary

A summary data frame comprising of the topic numbers, labels, coherence scores, prevalence scores, and top terms.

Examples

# \donttest{
# Create LDA Topic Model 
save_dir_temp <- tempfile()
dtm <- topicsDtm(data = dep_wor_data$Depphrase)

model <- topicsModel(
dtm = dtm, # output of topicsDtm()
num_topics = 20,
num_top_words = 10,
num_iterations = 1000,
seed = 42)
# }