Supervised topic model generating different topics to training data about bertopic

[MaartenGr]

Hmmm, it seems that the topics are correctly assigned but that they are re-ordered internally to make sure that the more frequent topics get lower values (they are sorted I think). So no worries, they are not actually different classifications just merely different IDs for the topics. This means that you can do one of two things. First, you can map training_topic_number to model_topic_number since, for instance, 4 always seems to map to 5. Second, I am not sure but I remember that manual BERTopic does not perform this sorting process.

[morrisseyj]

@MaartenGr thanks for the response. I had already sorted the data by count when assigning the topic numbers, so that couldn't have been the problem. Going back through and checking it turns out there was an error on my side with the way i had amalgamated topics earlier on. You explanation made me check more closely and catch this. Apologies for missing this the first time around and wasting your time. I really appreciate the library you have put together here.

[MaartenGr]

No problem! I'm just glad that your problem was resolved and thanks for the kind words 😄