In summary, we were granted access to the University of Vermont's StoryWrangler by the Computational Stories Lab which, among other things, randomly subsamples 10% of all tweets to evaluate collective emotion.
We collected 110 368 English tweets containing the word “Jew” (01 Sep – 01 Nov 2022) to study online discourse before and during Kanye West’s antisemitic outbursts.
Two DistilBERT classifiers were trained with HuggingFace Transformers:
- Our custom model – 700 tweets hand-labeled (IHRA definition) by four annotators that we trained and employed on our dataset
- institute for the study of contemporary antisemitism “gold-standard” model – 6 000 pre-annotated tweets from Jikeli et al.
The models and dataset can be found on my huggingface account: https://huggingface.co/mivry (30+ downloads last month alone in huggingface)
Here are some interesting graphics that coincide with Kanye's antisemitic tweet on October 8th 2022:
And Kanye's interview on October 16th 2022, where he urges, among other things, to "#MeToo-ing the Jewish culture. I’m saying y’all gotta stand up and admit to what you been doing" See:
https://www.youtube.com/watch?v=7IavRTSxvRo
The paper itself is buried in my folders. When I make a blog I may post it there more casually :)