metadata

license: mit
language:
  - en
library_name: transformers
pipeline_tag: token-classification
tags:
  - Social Bias
metrics:
  - name: F1
    type: F1
    value: 0.7864
  - name: Recall
    type: Recall
    value: 0.7617
base_model: bert-base-uncased
co2_eq_emissions:
  emissions: 8
  training_type: fine-tuning
  geographical_location: Phoenix, AZ
  hardware_used: T4

Social Bias NER

This NER model is fine-tuned from BERT, for multi-label token classification of:

(GEN)eralizations
(UNFAIR)ness
(STEREO)types

You can try it out in spaces :).

How to Get Started with the Model

Transformers pipeline doesn't have a class for multi-label token classification, but you can use this code to load the model, and run it, and format the output.

import json
import torch
from transformers import BertTokenizerFast, BertForTokenClassification
import gradio as gr

# init important things
tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased')
model = BertForTokenClassification.from_pretrained('ethical-spectacle/social-bias-ner')
model.eval()
model.to('cuda' if torch.cuda.is_available() else 'cpu')

# ids to labels we want to display
id2label = {
    0: 'O',
    1: 'B-STEREO',
    2: 'I-STEREO',
    3: 'B-GEN',
    4: 'I-GEN',
    5: 'B-UNFAIR',
    6: 'I-UNFAIR'
}

# predict function you'll want to use if using in your own code
def predict_ner_tags(sentence):
    inputs = tokenizer(sentence, return_tensors="pt", padding=True, truncation=True, max_length=128)
    input_ids = inputs['input_ids'].to(model.device)
    attention_mask = inputs['attention_mask'].to(model.device)

    with torch.no_grad():
        outputs = model(input_ids=input_ids, attention_mask=attention_mask)
        logits = outputs.logits
        probabilities = torch.sigmoid(logits)
        predicted_labels = (probabilities > 0.5).int() # remember to try your own threshold

    result = []
    tokens = tokenizer.convert_ids_to_tokens(input_ids[0])
    for i, token in enumerate(tokens):
        if token not in tokenizer.all_special_tokens:
            label_indices = (predicted_labels[0][i] == 1).nonzero(as_tuple=False).squeeze(-1)
            labels = [id2label[idx.item()] for idx in label_indices] if label_indices.numel() > 0 else ['O']
            result.append({"token": token, "labels": labels})

    return json.dumps(result, indent=4)

GUS-Net Project Details:

Resources:

Please visit this collection for the datasets and model presented in the GUS-Net paper.
GUS-Net was implemented as part of The Fair-ly Project, in a Chrome Extension, and PyPI package.

Please cite:

@article{powers2024gusnet,
  title={{GUS-Net: Social Bias Classification in Text with Generalizations, Unfairness, and Stereotypes}},
  author={Maximus Powers and Umang Mavani and Harshitha Reddy Jonala and Ansh Tiwari and Hua Wei},
  journal={arXiv preprint arXiv:2410.08388},
  year={2024},
  url={https://arxiv.org/abs/2410.08388}
}

Give our research group, Ethical Spectacle, a follow ;).