summarization_amr_graph

This is the re-implementation for the paper : Toward Abstractive Summarization Using Semantic Representations

We will note our develop step in next few months.

Step1

1. Transform the origin CNN/Daily Mail dataset into AMR format.

   This part will follow the paper Automatic Lossless-Summarization and its' repository.

2. the problem is that Lossless-ATS do not use the state-of-the-art models to parse text, and alse use python2.x, we will refine the code in the brach cnndm_to_amr

Step2

1. In this step, we use allennlp and DGL to re-implement the graph-to-sequence model mentioned in Structured Neural Summarization
2. The code is avaliable in branch structured_model

Step3

1. graph-to-graph for Molecular
2. internal-graph for text