A single-label classifier for universal framing dimensions in news articles on political topics. These dimensions and their respective short descriptions are:
- Economic: Political conflict around the role of the state in the economy
- Cultural: Political conflict of individualism vs traditionalism
- Governance:Pluralist vs. populist and technocratic understanding of the society and political conflict
- Affective: Distinction of political in- and out-groups invoking positive or negative emotions
We are currently working on a paper detailing these dimensions and their conceptualization in more detail.
Create python environment, for example with conda. Python 3.8 or later is supported.
conda create --yes -n NewsFrames python=3.8
conda activate NewsFramesInstall:
pip install NewsFramesfrom NewsFrames import Classifier
classifier = Classifier()
results = classifier.predict(["Executives at the British software company Autonomy mischaracterised revenues from clients including Tottenham Hotspur, the Serious Fraud Office and the BBC to inflate software sales figures before a disastrous £8bn acquisition by the US firm Hewlett-Packard, London’s high court has heard."])
print(results)You can use the attribute_mode parameter to get predictions for the individual
attributes (attribute_mode="withattributes") or only whether the respective dimension
is present or not (attribute_mode="withoutattributes"). The default is
withattributes.
classifier = Classifier(attribute_mode="withoutattributes")Thanks to Tilman Hornung for preparing and analyzing the datasets the NewsFrames classifier was trained and evaluated on.
python -m pip install build twine
python -m build
python -m twine upload dist/*
rm -rf dist/*