Calculate NPMI for Topic Top-Terms

NPMI (Normalized Pointwise Mutual Information) measures coherence of top-K terms per topic using internal document co-occurrence from the supplied DFM (boolean presence).

Usage

calculate_npmi(top_terms_list, dfm, top_k = 10, epsilon = 1e-12)

Arguments

top_terms_list

Named or unnamed list of character vectors, one per topic.

dfm

A quanteda dfm providing the reference doc-term frequencies.

top_k

Number of top terms per topic to include (default 10).

epsilon

Numerical stabilizer (default 1e-12).

Value

list with mean_npmi (scalar) and per_topic_npmi (numeric vector).

See also

Other topic-modeling: analyze_semantic_evolution(), assess_embedding_stability(), assess_hybrid_stability(), auto_tune_embedding_topics(), calculate_assignment_consistency(), calculate_keyword_stability(), calculate_semantic_drift(), calculate_topic_diversity(), calculate_topic_probability(), calculate_topic_stability(), extract_topic_terms_df(), find_optimal_k(), find_topic_matches(), fit_embedding_model(), fit_hybrid_model(), fit_temporal_model(), fit_topic_prevalence_model(), generate_topic_labels(), get_topic_prevalence(), get_topic_terms(), get_topic_texts(), identify_topic_trends(), plot_model_comparison(), plot_quality_metrics(), plot_topic_effects_categorical(), plot_topic_effects_continuous(), plot_topic_probability(), plot_word_probability(), run_neural_topics_internal(), validate_semantic_coherence()