Datasets:

schema_guided_dstc8

like 8

Hi, I am currently working on action prediction with this dataset. In theory, it should not be a complex problem but I have doubts because looking at the dataset I have found several cases where for the same state (I understand as state a combination of user intentions, slots and previous actions) we have several possible actions and also for the same action we have different next states. Therefore a dialogue would have a stochastic factor and if it is due to randomness in the labelling it is very complex for a model to capture this pattern.

PD:
I am working with Rasa's Transformer Embedding Dialogue (TED). Using only supervised learning.
https://arxiv.org/abs/1910.00486