Unior-NLP @PAN2020_Profiling-Fake-News-Spreaders

UniOR NLP Group code repository for the task at PAN-CLEF 2020 "Profiling Fake News Spreaders".

In that workshop, UniOR NLP Group participated in both English and Spanish subtasks. We used different machine learning algorithms combined with strictly stylometric features, categories of emojis and a bunch of lexical features related to the fake news headlines vocabulary.

Here you can find the code used and the submitted models.

Data

For a complete description of both data and task, please refer to the Task Website or Zenodo.

How to use

The commands below show how to use the scripts.

The train_fnsp.py script trains the English and Spanish models using the data located at input data directory and saves the trained models to output directory path.

python3 train_fnsp.py -i "input data directory" -o "output directory path"

The predict_fnsp.py scripts read the data from an input data directory and write to an output directory the predicted labels for each document of the dataset.

python3 predict_fnsp.py -i "input directory path" -o "output directory path"

Citing

@InProceedings{manna:2020,
  author =              {Raffaele Manna and Antonio Pascucci and Johanna Monti},
  booktitle =           {{CLEF 2020 Labs and Workshops, Notebook Papers}},
  crossref =            {pan:2020},
  editor =              {Linda Cappellato and Carsten Eickhoff and Nicola Ferro and Aur{\'e}lie N{\'e}v{\'e}ol},
  month =               sep,
  publisher =           {CEUR-WS.org},
  title =               {{Profiling Fake News Spreaders Through Stylometry and Lexical Features: UniOR NLP @PAN2020---Notebook for PAN at CLEF 2020}},
  url =                 {http://ceur-ws.org/Vol-2696/},
  year =                2020
}