Skip to contents

Embedding-Based Topic Discovery

Source: R/topic_modeling.R
cluster_embedding_topics.Rd

Groups documents into topics by clustering transformer embeddings, with per-topic cohesion diagnostics.

Usage

cluster_embedding_topics(
  texts,
  n_topics = 10,
  embedding_model = "all-MiniLM-L6-v2",
  clustering_method = c("kmeans", "hierarchical"),
  min_topic_size = 3,
  seed = 123
)

Arguments

texts

Character vector of documents

n_topics

Number of topics to discover

embedding_model

Transformer model for initial embeddings

clustering_method

Algorithm applied to the embedding similarity matrix: "kmeans" (default) or "hierarchical". Both honour n_topics and assign every document.

min_topic_size

Minimum documents per topic.

seed

Random seed for reproducibility

Value

List with topic assignments and diagnostics. The cohesion values are mean pairwise cosine similarity of document embeddings within each topic cluster (embedding-space compactness), not lexical coherence measures such as C_v or NPMI.

See also

find_optimal_k() and auto_tune_embedding_topics() for choosing n_topics; fit_embedding_model() for the UMAP and HDBSCAN pipeline that derives the topic count from density instead.