NLP notebooks in kaggle

Hi All,

I’m starting a serie of notebooks based in the NLP specialisation in kaggle, adapting the knowledge acquired to competitions or datasets in kaggle.

I will be really happy to discuss and improve them with your comments.
This notebook is based in the first week of the first course in the specialisation. I solved the Getting Started competition: ‘Natural Language Processing with Disaster Tweets’ scoring 0.74134. I think thats maybe it’s possible to improve this score using this technique, and presented some ideas how to do it at the end of the notebook.
This is not directed based in any course, but I get inspired by the third course (that I’m still not enrolled). I tried a simple approach to Transfer Learnig using Universal Sentence Encoder. The score obtained si much better than with the first notebook 0.81305. I have some ideas how to improve it, and fighting with the overfitting, and hoping that the thirst course will give me some ideas ho to solve it in a best way!

Please, feel free to use the notebooks, fork, copy it… and if you have some comments, please don’t doubt to use the comment sections in kaggle, or just reply to this post :slight_smile:

May the NLP be with you :wink:


Hi Pere_Martra,

Very cool!

The fourth course on transformers may give you some additional tools. If you would like to try a shortcut, you could have a look at hugging face’s text classification models, (but you might want to work through the fourth course first).

Keep your posts coming!

1 Like