Stylized Story Generation with Style-Guided Planning

This repository contains the pytorch implementation of "Stylized Story Generation with Style-Guided Planning".

Xiangzhe Kong∗, Jialiang Huang∗, Ziquan Tung, Jian Guan and Minlie Huang†

Code description

• Code for extracting keywords and annotating the dataset are in data .
• Code for GPT2/Bart/Our models are in gpt2baseline / bartbaseline / ours respectively.
• Code for calculating BLEU / SSC / LSC are in analyze.

Run

We provide a shell script that incorporate both training and testing procedure.

• To train four models (GPT2, Bart, Ours, Bert for SSC) at the same time (4 GPUs are needed)

bash run.sh parallel

• To train four models sequentially, using only one GPU

bash run.sh serial

Logs about the training and testing will be saved in the directory of each model respectively.

Besides, BLEU scores are saved in analyze/bleu and LSC/SSC scores are saved in analyze/LSC_SSC.