Structure-Aware-BART

This repo contains codes for the following paper:

Jiaao Chen, Diyi Yang:Structure-Aware Abstractive Conversation Summarization via Discourse and Action Graphs, NAACL 2021

If you would like to refer to it, please cite the paper mentioned above.

Getting Started

These instructions will get you running the codes of Structure-Aware-Bart Conversation Summarization.

Requirements

• Python 3.6 or higher
• Pytorch >= 1.3.0
• Pandas, Numpy, Pickle
• rouge=1.0.0 (https://github.com/pltrdy/rouge)
• transformers
• allennlp
• openie
• wandb

Note that different versions of rouge or different rouge packages might result in different rouge scores. For the transformers, we used the version released by Oct. 7 2020. The updated version might also result in different performances.

Install the transformers with S-BART

cd transformers

pip install --editable ./

Downloading the data

Please download the SAMSUm dataset

Pre-processing the data

The data folder you download from the above link already contains all the pre-processed files (including the extracted graphs) from SAMSum corpus.

Extract Discourse Graphs

Here we utilize the data and codes from here to pre-train a conversation discourse parser and use that parser to extract discourse graphs in the SAMSum dataset.

Extract Action Graphs

Please go through ./src/data/extract_actions.ipynb to extract action graphs.

Training models

These section contains instructions for training the conversation summarizationmodels.

The generated summaries on test set for baseline BART and the S-BART is in the ./src/baseline and ./src/composit folder. (trained with seed 42)

The training logs from wandb for different seed (0,1,42) for S-BART is shown in ./src/Weights&Biases.pdf

Training baseline BART model

Please run ./train_base.sh to train the BART baseline models.

Training S-BART model

Please run ./train_multi_graph.sh to train the S-BART model.

Evaluating models

Please follow the example jupyter notebook (./src/eval.ipynb) is provided for evaluating the model on test set.