Model returns a CSV file of the post ID and the hazard probability for each of theses posts.
python hazard_detection.py <filename>
Lists the required libraries:
- pandas==2.0.2
- sklearn==1.2.2
- matplotlib==3.7.1
- scipy==1.10.1
- sentence_transformers==2.2.2
- numpy==1.24.4
This contains all code to have GPT annotate tweets or urban legends, as well as summarize all urban legends to under 128 tokens.
This contains all code and guidelines for collection and annotation of ground truth X posts, along with the annotated posts and urban legend annotations from Fessler et al., 2014. We also share a datasheet that describes the data.
This contains the trained model and the code to predict hazards in JSONL files.
This contains the coordinated network extracted from our analysis of X posts related to the Hamas-Israel war in an edgelist format. The nodes represent hashed usernames.
This contains the Jupyter notebook used to train the models
This contains all urban legends collected, including those from Fessler et al., 2014. We also share a datasheet that describes the data.
References:
Fessler DMT, Pisor AC, Navarrete CD (2014) Negatively-Biased Credulity and the Cultural Evolution of Beliefs. PLoS ONE 9(4): e95167. doi:10.1371/journal.pone.0095167