From what I can see the issue of duplicates happens with manual scrolling, and not if I use an extension to automatise the process.
When present, duplicates consist of the same tiktok videos appearing 2 times, with the same metadata. In the datasets I collected (around 4.000-5.000 posts), after deleting the duplicates the datasets decreased, on average, of 1000 posts.