Look into:
- Sentence folder: for the sentence encoder based model (Recommended)
- Paragraph folder: for the sentence and paragraph encoder based model
- A BiLSTM Encoder for the sentence
- A BiLSTM Encoder for the paragraph
- Combination (Concatenation) of the final states of the two encoders act as the intial state for decoder
- Several smaller tweaks to this
- Uses ONMT library
Implementation: PyTorch
Ideas from: https://arxiv.org/abs/1705.00106
A report of the several different directions taken and the experiments is as follows:
Report: http://www.cse.iitd.ac.in/nlpdemo/ques_gen/report.pdf
Demo version has been hosted at: http://www.cse.iitd.ac.in/nlpdemo/ques_gen
@Prakhar0409
@ArkaSaha