Entropy based Attention Regularization 👂 This is an English BERT fine tuned with Entropy based Attention Regularization to reduce lexical overfitting to specific words on the task of Misogyny Identification. Use this model if you want a debiased alternative to a BERT classifier. Please refer to the paper to know all the training details. Dataset The model was fine tuned on the Automatic Misogyny Identification dataset. Model This model is the fine tuned version of the bert base uncased model. We trained a total of three versions for Italian and English. Model Download bert base uncased ear misogyny Link bert base uncased ear mlma Link bert base uncased ear misogyny italian Link Authors Giuseppe Attanasio Debora Nozza Dirk Hovy Elena Baralis Citation Please use the following BibTeX entry if you use this model in your project: Limitations Entropy Attention Regularization mitigates lexical overfitting but does not completely remove it. We expect the model still to show biases, e.g., peculiar keywords that induce a specific prediction regardless of the context. Please refer to our paper for a quantitative evaluation of this mitigation. License GNU GPLv3
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy