@srchvrs It was never disclosed what the subsample was, but I think it was more like 1%. Now you could of course discard all the metadata and just stick with the text of the tweet, but then you‘d be discarding also a lot of data that could help you identify bots.