colinsongf / feature-rich-pointer-generator

An extension for pointer-generator network in Abstractive Text Summerization.

Feature Rich Pointer Generator

This is a MLP group project work in University of Edinburgh. Our work is an extension for the pointer-generator network proposed by See (2017). See the report in the folder docs for detailed information.

Usage

Step 1. install pacakge

git clone https://github.com/JZ95/feature-rich-pointer-generator
cd feature-rich-pointer-generator
# you may create a new virutal env for that
python setup.py develop
# run the command to see the help info
frpg_run --help

Step 2. get CNN/Daily Mail Data Set

you may see the instructions here and preprocess the data on you own, or just use the data processed by us by going through the following steps.

1. Download the tarball of data from Google Drive.
2. decompress the tarball tar -xzf data.tar.gz, then get a folder named data.
3. move the data folder into the ROOT path of this repo.

Step 3. run your experiment

check the SHELL scripts in the folder scripts, see the comments for detailed explaination.

---

TIPS for Setting pyrouge.

1. install python wrapper for ROUGE

pip install pyrouge

2. set rouge path clone the ROUGE PERL repo

git clone https://github.com/andersjo/pyrouge
pyrouge_set_rouge_path /absolute/path/to/pyrouge/tools/ROUGE-1.5.5

3. run evaluation

frpg_eval_rouge --dir=/path/to/your/result

Docker

Refer the file docs/docker_usage.md for guidelines to deploy this repo using docker.

---

Collaborators:

@Shihao Liu, @Christos Drou