How to link data inside topic model to original training data without preprocessed?

Nobody has claimed this yet.

I use bertopic in Chinese .
So I must split the text to token and add space between every 2 tokens, and remove stopwords, and some special symbol, than I input the processed text into bertopic.
And the problem is how can I link the document to the oringinal training data without preprocessed? if I use documents in the model, the data are tokens without stopwords etc., its s weired.
except transform every text again？Somebody get better ideas?

No source file, test, or entry point is named. Start by tracing how BERTopic accepts processed documents and exposes documents in its topic model, then determine whether original-text mapping can be supported without reprocessing. Done means the relationship between processed and original training data is preserved and verified for the described Chinese-tokenization workflow.

Written by the indexing model from the issue text.