EventGround: Narrative Reasoning by Grounding to Eventuality-centric Knowledge Graphs

This repo contains the sample code for reproducing the results of our LREC-COLING'24 paper: EventGround: Narrative Reasoning by Grounding to Eventuality-centric Knowledge Graphs.

Requirements

Install the following packages:

dgl==0.8.1
pytorch==1.9.1
allennlp==2.9.3
transformers==4.21.0
datasets==2.0.0
sentence-transformers==2.2.2
faiss-gpu==1.7.2
wandb==0.12.14

Data

Start by downloading the KG data and the related datasets.

./data

This directory is for storing ASER KGs, embeddings, Faiss index, and so on.

Download aser data to ./data from here: https://hkust-knowcomp.github.io/ASER

./dataset

This directory is for storing narrative reasoning datasets.

Preprocessing

To set up the eventuality retriever, first embed ASER nodes with retrieval_pipeline/get_aser_event_embeds.py. Then, train a Faiss accelerator with retrieval_pipeline/train_faiss.py.

With the retriever prepared, run the preprocessing scripts in the following order to obtain grounded eventuality subgraphs.

1. preprocess_[DATASET]_events.py for event extraction.
2. preprocess_[DATASET]_pairs.py to find event anchors for the extracted events.
3. preprocess_[DATASET]_sp.py to find shortest paths on eventuality KGs.
4. preprocess_[DATASET]_graphs.py to construct subgraphs.

Training

Refer to the run.py scripts under dataset specific folders (SCT, MCNC). All the training and evaluation results will be found in the wandb panel.

Misc

If you use this research, please cite us:

@article{jiayang2024eventground,
  title={EventGround: Narrative Reasoning by Grounding to Eventuality-centric Knowledge Graphs},
  author={Jiayang, Cheng and Qiu, Lin and Chan, Chunkit and Liu, Xin and Song, Yangqiu and Zhang, Zheng},
  journal={arXiv preprint arXiv:2404.00209},
  year={2024}
}

If you have any questions, please send an email to [email protected].