Skip to content

ganesh25/Joke-generator

Repository files navigation

Joke-generator

Alexa Skill to read Jokes generated using a Deep Learning model.

Data

  • We created a data crawler script to scrap jokes from reddit.

Pipeline

PipelineImage

Models used

  • We initially used a LSTM model for training, but this didn't yield good jokes.

  • Then we decided to use GPT-2, since it is already trained on a huge corpus.

  • As a result, our new model generated better jokes.

Alexa Skills Interface with Examples

Example1

Example2

Future Scope

  • Remove offensive jokes due to biased data
  • 1 forward pass ~ 45 seconds
  • Make the model fetching + reading out real time and automated
  • Make the model learn the difference between Humour and Sarcasm, since they are both distinct

@misc{pungas, title={A dataset of English plaintext jokes.},
url={https://github.com/taivop/joke-dataset},
author={Pungas, Taivo},
year={2017},
publisher = {GitHub},
journal = {GitHub repository}
}

About

Applied Deep Learning for NLP seminar

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published