Predicting new documents

To classify a new document into one of the topics, we just need to use .transform(). As discussed earlier, the difference between .fit_transform() and .transform() is that .fit_transform() trains and predicts documents, whereas .transform() predicts documents. Let’s load the test dataset:

test = pd.read_csv(path + “/ag_news_test.csv”)test_docs = test['Description']

We will load the model that has been trained by using the .load() function:

from bertopic import BERTopicmodel = BERTopic.load(path + “/ag_news_bertopic”)

It is just that simple:

predicted_topics, predicted_probs =    model.transform(test_docs)
print(predicted_topics)

The output is as follows:

[1, 6, 3, -1, 3, 15, -1, -1, 100, 100, 68, 12, 12, 3, 3, 3, 6, 6, 3, 10, 25, -1,…]

BERTopic adopts a modular design approach that consists of five modules. You can choose techniques other than the default techniques...