query-wellformedness
query-wellformedness copied to clipboard
google-research-datasets
Training a model using this dataset
I want to train a model using this dataset so that when I feed it with grammatically incorrect questions it should rate it. How can create such model?
Well, that's the thing you have to find out. And now that you have the labeled data you can try out which NN-architecure performs best on that task. The hope is that it finds out by itself which one is syntactically wrong. I would even consider looking for grammar checkers, like in MS Word or grammarly; perphas like this you come up with some creative ideas to improve SOTA deep learning approaches. And you might wanna use sth. like Universal Sentence Encoder to create some kind of embeddings. Just search the web and combine what you find — be creative :-)
Edit: Here is what you want (the associated paper) Identifying Well-formed Natural Language Questions.
@MartinLichtblau I tried training this using neural networks with different hyperparameter setting but, I am not able to achieve accuracy above 30%. I used Universal Sentence Encoder for embedding the dataset. Can you provide some help?
