metadata

library_name: transformers
datasets:
  - vector-institute/newsmediabias-plus
language:
  - en
metrics:
  - accuracy
  - precision
  - recall
  - f1
base_model:
  - google-bert/bert-base-uncased
pipeline_tag: text-classification

BERT NMB+ (Disinformation Sequence Classification):

Classifies sentences as "Likely" or "Unlikely" biased/disinformation (max token len 128).

Fine-tuned BERT (bert-base-uncased) on the headline and text_label fields in the News Media Bias Plus Dataset.

This model was trained with weighted sampling so that each batch contains 50% 'Likely' examples and 50% 'Unlikely' examples. The same model trained without weighted sampling is here, and got slightly better eval metrics. However, this model preformed better when predictions were evaluated by gpt-4o as a judge.

Metics

Evaluated on a 0.1 random sample of the NMB+ dataset, unseen during training

Accuracy: 0.6745
Precision: 0.9070
Recall: 0.6288
F1 Score: 0.7427

How to Use:

from transformers import pipeline

classifier = pipeline("text-classification", model="maximuspowers/nmbp-bert-headlines-balanced")
result = classifier("He was a terrible politician.", top_k=2)

Example Response:

[
  {
    'label': 'Likely',
    'score': 0.9967995882034302
  },
  {
    'label': 'Unlikely',
    'score': 0.003200419945642352
  }
]