topic_model.py

def get_review_topic(review_text, vectorizer, nmf_model, top_n_words=5):
    # Transform the review into TF-IDF space
    review_vect = vectorizer.transform([review_text])
    
    # Get topic distribution
    topic_probs = nmf_model.transform(review_vect)  # shape (1, num_topics)
    topic_index = topic_probs.argmax()              # pick topic with highest score
    
    # Get top words for that topic
    words = vectorizer.get_feature_names_out()
    topic_words = [words[j] for j in nmf_model.components_[topic_index].argsort()[-top_n_words:][::-1]]
    
    return ", ".join(topic_words)