This repo has code for our paper "Breaking the Narrative: Scene Segmentation through Sequential Sentence Classification" that ranked 1st at the Shared Task on Scene Segmentation (STSS).
The easist way to run our model on your data is through Docker. Please put the novels you want segmented into the "data/test" directory following the shared task's JSON format.
Then you can simply build & run the docker image by simply running:
./build.sh
./run.sh
The built image will be ~9 GB and the predictions will be saved in "predictions" folder: One output file will be created per input file with the same name.
If you use the data or the model, please cite,
@inproceedings{kurfali2021breaking,
title={Breaking the Narrative: Scene Segmentation through Sequential Sentence Classification},
author={Kurfal{\i}, Murathan and Wir{\'e}n, Mats}
journal={Shared Task on Scene Segmentation},
year={2021}
}