Modeling Protagonist Emotions for Emotion-Aware Storytelling

This repository contains preliminary code and data for the paper titled:

Modeling Protagonist Emotions for Emotion-Aware Storytelling Faeze Brahman, and Snigdha Chaturvedi. EMNLP 2020.

Dataset: ROCStories

The dataset can be downloaded from here and unzipped in data/ folder.

Data files includes:

1. [train/test/dev]_x1.txt: titles
2. [train/test/dev]_x4.txt: stories
3. [train/test/dev]_mapped.txt: automatically annotated emotion arcs

Code

• The code depends on Texar. Please install the version under third_party/texar. Follow the installation instructions in the README there.
• Download gpt-2-M from here and put it in gpt2_pretrained_models/ folder.
• The BERT-based classifier is trained using fast-bert. Please git clone (or pip install) it and use run_classifier_bert.py to train the emotion classifier.
• For obtaining emotional reactions, please git clone COMET here. And move comet_generate.py and find_x_o_appx.py there.
• Use prepare_data.py to preprocess the story data and transform them into TFRecord format. An example command is (please see the code for more config options).

python prepare_data.py --data_dir=data

• Run run_[X].sh for training/testing model [X]. (please see config files for more config options.)
• Use Reinforcement/run_evaluation.py for evaluation on emotion faithfulness. An example command is:

python Reinforcement/run_evaluation.py --all-preds-dir <PATH_TO_GENERATED_TSV_FILE> --arc-file <PATH_TO_ARC_FILE>  --output_file <PATH_TO_SAVE_JSON_RESULTS>

• BLEU scores measurements:

perl LIB/multi-bleu.perl data/test_x4.txt < <PATH_TO_GENERATED_TXT_FILE>

• The Distinct-n scores in the paper use the code here.

Interactive Generation

First, download the pretrained model from here and untar it:

tar -xvzf model_checkpoint.tar.gz

Then run following command to interactively generate emotion-aware stories:

sh run_interactive.sh

Running that, it will ask you to first enter a Title, and then a sequence of three emotions separated by space from joy, anger, sadness, fear, neutral! for example: joy sadness sadness

The code is adapted from Counterfactual Story Generation.

Reference

Please cite our paper using the following bibtex:

@inproceedings{brahman-chaturvedi-2020-modeling,
    title = "Modeling Protagonist Emotions for Emotion-Aware Storytelling",
    author = "Brahman, Faeze  and
      Chaturvedi, Snigdha",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.emnlp-main.426",
    pages = "5277--5294"
}