MLGCN-DP for Story Ending Generation

Code for paper "Story Ending Generation with Multi-Level Graph Convolutional Networks over Dependency Trees", AAAI 2021.

Prerequisites

• Python 3.6
• PyTorch == 1.14.0

Quick start:

• Dataset

All dataset under the directory of data, include ROCStories corpus (data/test, data/train, data/val) and the dependency parsing relations (data/DP-data.zip, unzip it to use). We utilize the glove embedding, please download the glove.6b.300d.txt and put it in directory of data/embedding.

• Data preprocess

Run following command:

1. python data_preprocess.py
2. python embed_vocab.py

Then we will get three files

1. data/final_gcn_data.pt
2. data/embedding/embedding_enc.pt
3. data/embedding/embedding_dec.pt

• Training

Run command:

python train.py -gpu 0

• Inference

Run command:

python generate_story.py -gpu 0

Then the output file will be save in directory of story_generation/

PS

When calculate the BLEU, please uniform case of ref and hyp