DeepPavlov
DeepPavlov copied to clipboard
Creating intents for go-bot using small number of seed utterances.
The biggest issue I am facing in creation of training data for go-bot is creation of new intents. Here is something I would like to suggest:-
1.) Apply Sequence to Sequence with attention model on SemEval dataset.
2.)Loop through each seed sentence and create a new sentence using model created in 1.).
3.) Take the generated sentences as seed utterances and stop doing it if they are not meaning the same using Siemese networks.
Warm Regards,