This repository contains code and data for the paper:

Luo, Y., Card, D. and Jurafsky, D. (2020). Detecting Stance in Media on Global Warming. In Findings of the Association for Computational Linguistics: EMNLP 2020.

BibTex TBA

1. Create and activate a Python 3.6 environment.
2. Run pip install -r requirements.txt.
3. Re-install neuralcoref with the --no-binary option:

pip uninstall neuralcoref
pip install neuralcoref --no-binary neuralcoref

4. Download SpaCy's English model: python -m spacy download en
5. Update the config.json file with your local OS variables.

• Our dataset GWSD itself can be accessed via GWSD.tsv in the main directory. The dataset contains tab-separated fields for each of the following:
  1. sentence: the sentence
  2. worker_0, ..., worker_7: ratings from each of the 8 workers for the stance of the sentence
  3. disagree: the probability that the sentence expresses disagreement with the target (that climate change/global warming is a serious concern), as estimated by our Bayesian model
  4. agree: ditto for the "agrees" label
  5. neutral: ditto for the "neutral" label
  6. guid: a unique ID for each sentence
  7. in_held_out_test: whether the sentence was used in our held-out-test set for model and baseline evaluation

Note: The first 5 rows are the 5 screen sentences we use to make sure that annotators correctly understand the task, and thus do not have estimated probability distributions.

• Our lexicons of framing devices are located in 4_analyses/lexicons.
• The sequence of code to replicate our results can be found in the individual READMEs of the numbered sub-directories.