A further Breakdown of the Kindly roadmap #57
The current implementation of the kindly repository employs a very helpful pre-trained model called cardiffnlp/twitter-roberta-base-offensive. This is trained with over 50 million tweets to determine offensiveness and sentiment.
This and other open-sourced pre-trained models inspire how Kindly's model will also be shaped.
[x] Build basic ML model using regular ML steps #77
[x] Test model with initial data in our initial dataset and provide feedback
[-] Write retrained model using cardiffnlp/twitter-roberta-base-offensive
[ ] Replace current model in API with the retrained model (if data points is a lot eg 1000-5000 phrases labelled offensive / not-offensive)
[ ] Compare and benchmark the accuracy of both models.
- This would be the litmus test for our initial model in step 1
[ ] Replace Cardiff NLP model in API with the initial model (needs an official name)
A further Breakdown of the Kindly roadmap #57
The current implementation of the kindly repository employs a very helpful pre-trained model called cardiffnlp/twitter-roberta-base-offensive. This is trained with over 50 million tweets to determine offensiveness and sentiment.
This and other open-sourced pre-trained models inspire how Kindly's model will also be shaped.
[x] Build basic ML model using regular ML steps #77
[x] Test model with initial data in our initial dataset and provide feedback
[-] Write retrained model using cardiffnlp/twitter-roberta-base-offensive
[ ] Replace current model in API with the retrained model (if data points is a lot eg 1000-5000 phrases labelled offensive / not-offensive)
[ ] Compare and benchmark the accuracy of both models.
[ ] Replace Cardiff NLP model in API with the initial model (needs an official name)