TF(-IDF) over LSTM

I've just came across this package and noticed you used a Term Frequency approach over say (Tree-)LSTMs when the latter is apparently way better (https://www.mdpi.com/2078-2489/10/4/127, https://link.springer.com/chapter/10.1007%2F978-3-030-32233-5_46) (genuinely curious)?

Also, wouldn't using the full TF-IDF method be better than just TF? Or would you rather consider LDA?