Skip to contents

Generate Topic Labels with LLM Assistance

Source: R/langgraph_workflows.R
generate_topic_labels_langgraph.Rd

Uses LangGraph workflow to generate multiple label candidates for topics using local LLM (Ollama). Provides human-in-the-loop review of suggestions.

Usage

generate_topic_labels_langgraph(
  topic_terms,
  num_topics,
  ollama_model = "llama3",
  ollama_base_url = "http://localhost:11434",
  envname = "textanalysisr-env"
)

Arguments

topic_terms

List of character vectors, where each vector contains the top terms for a topic (from STM or other topic model)

num_topics

Integer, number of topics

ollama_model

Character string, name of Ollama model to use (default: "llama3")

ollama_base_url

Character string, base URL for Ollama API (default: "http://localhost:11434")

envname

Character string, name of Python virtual environment (default: "langgraph-env")

Value

List with:

  • success: Logical, TRUE if workflow completed successfully

  • label_candidates: List of label candidate objects for each topic

  • validation_metrics: Validation metrics (if available)

  • error: Error message (if failed)

Details

This function:

  1. Initializes LangGraph Python environment

  2. Calls Python workflow to generate label candidates

  3. Returns structured results for display in Shiny UI

  4. Allows human review and selection of labels

The workflow uses a StateGraph with nodes for:

  • Label generation (LLM)

  • Validation (LLM)

  • Conditional revision based on quality metrics

Examples

if (FALSE) { # \dontrun{
topic_terms <- list(
  c("education", "student", "learning", "teacher", "school"),
  c("health", "medical", "patient", "doctor", "treatment"),
  c("environment", "climate", "carbon", "emissions", "energy")
)

result <- generate_topic_labels_langgraph(
  topic_terms = topic_terms,
  num_topics = 3,
  ollama_model = "llama3"
)

if (result$success) {
  print(result$label_candidates)
}
} # }