Skip to content
Hand labeled 3% of ~650,000 tweets on vaccination stance and machine labeled the rest. A research on how the opinions show in retweet network communities.
Jupyter Notebook
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.ipynb_checkpoints
.DS_Store
SMSociety18_paper_72.docx
labeled_data.csv
modularity.png
paper.pdf
predict.ipynb
readme.md
stance.png
vaccination_train.ipynb

readme.md

This is a research project on exploring of online vaccination discussion and how it manifested in retweet network communities.

Three percent of the dataset is labeled as pro-vaccine, anti-vaccine, or neutral to vaccination. The labeled data is available in this repository and feel free to use it. The whole dataset is not included but available upon request.

These two pretty pictures are from this research as well:

How the data was labeled? In the labeled dataset we shared here, label "-1" means anti-vaccine sentiment, label "1" refers to pro-vaccine sentiment and "0" is neutral. We noticed that there's barely any truly "neutral" tweets in our dataset, we define neutral as tweets that do not show any leaning to either pro- or anti-vaccine sentiment. Two people labeled the data with 10 percent overlap. Before labeling, a criteria was established on how to define each labels. While positive and negative is more intuitive, they then agreed on defining neural tweets as those only contain facts, such as news reports or research results. The conflict labels from two annotators were solved by manual checking.

You can’t perform that action at this time.