Checking my understanding re: BERTopic 'topic_model.transform()'

[registjl]

Hi -- I'm a BERTopic newbie so please bear with me.

I want to confirm my understanding of the results from using 'topic_model.transform()' function.
Quick overview of my process: 

1. I've instantiated a BERTopic model using a pre-trained embedding model
   topic_model = BERTopic(umap_model=umap_model, hdbscan_model=hdbscan_model, calculate_probabilities=True, embedding_model=embedding_model)

2. I fit my topic_model with 'fit_transform' with my training data and save the model
   topics, probabilities = topic_model.fit_transform(TRAIN_DATA[text]) topic_model.save(model_name)

3. In another script I load the saved topic_model and use 'topic_model.transform' with new data.
   topic_model = BERTopic.load(topic_model_name) topics, probabilities = topic_model.transform(TEST_DATA)

MY QUESTION: The topics returned in step 3 are the "predicted" topics that each document in the TEST_DATA is expected to be part of. Is that right?

• In other words, there aren't any "new topics" created -- just assignment to one in the original model?

[MaartenGr]

MY QUESTION: The topics returned in step 3 are the "predicted" topics that each document in the TEST_DATA is expected to be part of. Is that right?

That is correct!

In other words, there aren't any "new topics" created -- just assignment to one in the original model?

Right again, no new topics are created with .transform.