On September 6th 2023 I participated in the ETHZ Latsis Symposium hackaton about detecting hatespeech online. I achieved third place and won 500 CHF.
I used an ensemble of these BERT models:
- xlm-roberta-base
- microsoft/mdeberta-v3-base
- aari1995/German_Semantic_STS_V2
- deepset/gelectra-base
- PM-AI/sts_paraphrase_xlm-roberta-base_de-en
- intfloat/multilingual-e5-large
- deutsche-telekom/gbert-large-paraphrase-euclidean
- ZurichNLP/swissbert
I also created a pre-training set using several older german hatespeech datasets.