Problem with essay scoring and lstm

Hello everyone, I’m using KNIME for a couple of month now and I realy like it. But now I’m stuck and have no idea where the problem is.
I’m trying to score german second language learner essays via a lstm aproach. For that I transformed the texts into there POS and feed them into the lstm.
The approach is inspired by an experiment in a paper witch tries the same for norweigian essays and the setup of the lstm similar to the lstm of the sentiment analysis exampel provided in KNIME Hub.

My Problem is, that I’m able to feed the POS transformed essays into the lstm but the network isn’t able to predict the lables. It isn’t learning at all.
Have anybody a idea what the Problem could be?
My exact workflow is:
pos_lstm.knwf (664.6 KB)

I hope somebody of you can help me with the problem, becaus I’m quite lost at the moment.

1 Like

Hi @EBastian -

Can you also provide some sample input data as well? Your preprocessing is quite specific to the format of the input data.

The Data that I use is the POS tagged Paula subcorpus of the MERLIN learner Corpus

The XMLs look like this:
1023_0001416_pos.xml (8.3 KB)

In any case, thanks a lot for your quick response

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.