Could you provicd the name of the operator where the process starts and never returns ?
Perhaps you may reduce the size of your select statement only using "title" ? If this works you
really need more RAM.
Why do you need operator "Nominal to Numerical" if TF-IDF delivers numerical values for all tokens found ?
And last but not least: Why you do not apply the "tokenize" operator inside "Prozess Documents" operator ?
You should start with tokenizing first and if this works you may add further operators like Generate-N-Grams and so on.