search

freesunshine0316 / MPQG

Code corresponding to our paper "Leveraging Context Information for Natural Question Generation"
47 stars 10 forks source link

Data logic #10

Open HarbingerOfHell opened 5 years ago

HarbingerOfHell commented 5 years ago

Hi, I'm trying to understand how text1, text2 and text3 are being derived in your experiment as there is no understanding of that in the paper. What is your logic for creating text1 and text3? Also what are the minimum requirements of text2?

freesunshine0316 commented 5 years ago

We had a description in the README file. text1, text2 and text3 correspond to the passage (or more specifically the sentence containing the answer), the target question and the answer, respectively.

HarbingerOfHell commented 5 years ago

Thanks for your prompt reply! I was actually wondering what your actual data source is or how you created it. How are you changing from the GloVe file to the vectors you are currently using?

freesunshine0316 commented 5 years ago

@HarbingerOfHell Hi, We follow previous work to use the SQuAD dataset for evaluation, and our embeddings are filtered according to our dataset. Those steps are standard.