The datasets are composed by Documents like PDFs and/or Office files?