This repository contains the dataset, experimental code and results presented in our EMNLP 2018.
Our dataset comprises 1860 short texts together with ratings for two kinds of empathic states, empathic concern and personal distress. It is, to our knowledge, the first publicly available gold standard for NLP-based empathy prediction. The csv
-formatted data can be found here. For details regarding our annotation methodology please refer to the paper.
Our dataset is available under CC BY 4.0.
We ran our code under Ubuntu 16.04.4. Our conda
environment is specified in environment.yaml
. To re-run our experiments, you have to add the root directory of the repository to you python path and setup an environment variable VECTORS
. Details can be found in the script activate_project_environment
and constants.py
. Before running the script, make sure that you have a properely named conda
environment set-up on your machine (default name is emnlp18empathy
).
Please note that re-running our code will produce varying results due to racing conditions caused by multi-threading.
The necessary FastText word vectors can be found here.
Once everything is set up, executing run_experiments.sh
will re-run our cross-validation experiment. The results will be stored in modeling/main/crossvalidation/results
.
@inproceedings{Buechel18emnlp,
author={Buechel, Sven and Buffone, Anneke and Slaff, Barry and Ungar, Lyle and Sedoc, Jo{\~{a}}o},
title = {Modeling Empathy and Distress in Reaction to News Stories},
year = {2018}
booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP 2018)}
}
You can find our arXiv preprint here.
I am happy to give additional information or get feedback about our work via email: [email protected]