Source code of competition in Fake-EmoReact 2021, the shared task of SocialNLP 2021 (in conjunction with NAACL 2021). We won the first place, and the report can be seen as here.
Given an unlabeled tweet and its GIF response, the model should predict the label of tweet as fake or real. The positive samples are tweets with the hashtag #FakeNews where the Covid related ones are included, and the negative samples are from the previous SocialNLP challenge EmotionGIF.
- train.json
- 168,521 samples with gold labels, to be used for training the model.
- dev.json
- 40,487 in Round 1.
- 45,036 in Round 2 (main track).
- eval.json
- 88,664 in Round 1.
- 110,492 in Round 2 (main track).
- The metric that will be used to evaluate entries are Macro-Precision, Macro-Recall, and Macro-F1 from Scikit-learn.
