Skip to content

The google-quest-challenge is an NLP competition that was hosted on Kaggle.

Notifications You must be signed in to change notification settings

arch-raven/google-quest-challenge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

google-quest-challenge

The google-quest-challenge is an NLP competition that was hosted on Kaggle. The task is to predict ratings for subjective aspects (like #question asker intent understanding ) of question answering on StackExchange Q&A pairs.

Link to Kaggle page

kaggle competitions download -c google-quest-challenge

Subjective Aspects

Things to try in priority order

  • Stable cross validation split, used GroupKFold with question_title group
  • Separate learning rate 3-e5 for transformers 0.005 for heads with cosine schedule
    • Freeze the transformer layer and train head for 1 epoch
    • train transformer layer and header layer for different learning rate
  • Using weighted sum of CLS outputs form all BERT layers rather than using only the last one, constraining the weights to be positive and sum to 1
  • Truncate the input text in different way when input is exceed the limit of input length
  • Different learning rate for each bert layer
  • Multi-Sample Dropout
  • Separate model for question and answer

About

The google-quest-challenge is an NLP competition that was hosted on Kaggle.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published