Heterogeneous Memory Network for Generative Question Answering
The implementation of paper: Yao Fu and Yansong Feng, Natural Answer Generation from Heterogeneous Memory, NAACL 2018
The blog of this paper can be found here.
The Cumulative Attention Mechanism is implemented in model.py (Actually it is just one more line of code, if you implement attention mechanism as a function) -- so this mechanism is highly recommended to improve your model performance.
In addition to the original QA dataset, it has been tested on many data-to-text generation tasks after the paper was published, and achieved very good performance.
The Neural Checklist model was originally implemented in Lua Torch, here we reimplemented in tensorflow.
(Sorry that the model is a bit of messy, this was my first experience in DL4NLP)