Binary
Classifying text sequences as biased/fair.
Overview of Task:
Binary classification is the foundation of many bias detection frameworks, and in this case refers to classifying an entire text sequence as "biased" or "unbiased."
This is typically implemented with an encoder-only model, such as BERT, to create encodings (i.e. contextual representations) that capture "the meaning" of a sentence, and can be passed to a classifier layer(s) with one output feature (for 0 to 1 probability of a single class: "Biased").
🤖 Models:
One of the UnBIAS findings is that ternary classification (see Multi-Class) is a stronger approach, but the binary classification model is just as good.
UnBIAS Classifier
UnBIAS is a framework started in 2023 by Raza. et al at the Vector Institute, and a refresh of the technology proposed in Dbias.
🤗 HF Space to Test UnBIAS Classifier
Base Model: bert-base-uncased Dataset: BEAD (3.67M rows)
🤗 Hugging Face Model
📄 Research Paper
Use UnBIAS Classifier:
# pip install transformers
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("newsmediabias/UnBIAS-classification-bert")
model = AutoModelForSequenceClassification.from_pretrained("newsmediabias/UnBIAS-classification-bert")
classifier = pipeline("text-classification", model=model, tokenizer=tokenizer , device=0 if device.type == "cuda" else -1)
classifier("Anyone can excel at coding.")Dbias
Dbias proposed an architecture in 2022, for addressing news media bias, with a framework that utilized binary classification, named-entity recognition, bias masking, and word recommendation (Raza, et al.).
While reimplementations have made changes in approach, Dbias was a trailblazer, especially for binary classification (the first phase in the image above).
Base Model: bert-base-uncased, Dataset: MBAD Dataset
🤗Hugging Face Model
📄 Research Paper
Use Dbias Bias Classification:
Dbias has a PyPI package.
# pip install Dbias
# pip install https://huggingface.co/d4data/en_pipeline/resolve/main/en_pipeline-any-py3-none-any.whl
from Dbias.bias_classification import *
# returns classification label for a given sentence fragment.
classifier("Tall people are so clumsy.")This is running list of cool binary classification models we've seen and want to learn more about. If you find one that should be here, send it to us on discord.
💾 Datasets:
Bias Evaluation Across Domains (BEADs) Dataset
3.67M rows | 2024
The BEADs corpus was gathered from the datasets: MBIC, Hyperpartisan news, Toxic comment classification, Jigsaw Unintended Bias, Age Bias, Multi-dimensional news (Ukraine), Social biases.
It was annotated by humans, then with semi-supervised learning, and finally human verified.
It's one of the largest and most up-to-date datasets for bias and toxicity classification, though it's currently private so you'll need to request access through HuggingFace.
🤗HuggingFace Dataset
📑 Contents
text
The sentence or sentence fragment.
dimension
Descriptive category of the text.
biased_words
A compilation of words regarded as biased.
aspect
Specific sub-topic within the main content.
label
Indicates the presence (True) or absence (False) of bias. The label is ternary - highly biased, slightly biased, and neutral.
toxicity
Indicates the presence (True) or absence (False) of toxicity.
identity_mention
Mention of any identity based on words match.
While BEADs doesn't have a binary label for bias, the ternary labels (e.g. neutral, slightly biased, and highly biased) of the label field can categorized into biased (1), or unbiased (0). Additionally, the toxicity field contains binary labels.
📄 Research Paper
Generalizations, Unfairness, and Stereotypes Dataset (Synthetic Corpus)
37.5k rows | 2024
The GUS dataset (released in the GUS-Net paper), is an entirely synthetic dataset. It was generated by Mistral 7B, and later used for named-entity recognition. The results of GUS-Net showed that the synthetic corpus was effective across domains, and less noise than authentic datasets.
🤗HuggingFace Dataset
📑 Contents
biased_text
The full text fragment where bias is detected.
racial
Binary label, presence (1) or absence (0) of racial bias.
religious
Binary label, presence (1) or absence (0) of religious bias.
gender
Binary label, presence (1) or absence (0) of gender bias.
age
Binary label, presence (1) or absence (0) of age bias.
nationality
Binary label, presence (1) or absence (0) of nationality bias.
sexuality
Binary label, presence (1) or absence (0) of sexuality bias.
socioeconomic
Binary label, presence (1) or absence (0) of socioeconomic bias.
educational
Binary label, presence (1) or absence (0) of educational bias.
disability
Binary label, presence (1) or absence (0) of disability bias.
political
Binary label, presence (1) or absence (0) of political bias.
sentiment
The sentiment given to Mistral 7B in the prompt.
target_group
The group Mistral7B was told to prompt.
statement_type
Type of bias prompted (e.g. "stereotypes," "discriminatory language," "false assumptions," "offensive language," "unfair generalizations").
Mistral 7B was prompted to generate biased sentences, using the arguments in the table above. This means the entire thing should be biased. You may want to supplement this with a dataset of fair statements, and label all the sentences from this dataset with 1 (biased).
📄 Research Paper
Bias Annotations By Experts (BABE)
4.12k records | 2023
Human annotated, and all annotators must agree. In its paper, BABE showed great results with BERT for sequence classification of news articles. While smaller than some other datasets, the annotations are very reliable (highly recommended as an external dataset for model eval).
🤗HuggingFace Dataset
📑 Contents
text
The text fragment (few sentences or less).
outlet
The source of the text fragments.
label
0 or 1 (biased or unbiased).
topic
The subject of the text fragment.
news_link
URL to the original source.
biased_words
Full words contributing to bias, in a list.
type
Political sentiment (if applicable).
📄 Research Paper
Not added yet
Not added yet
How it Works:
BERT (and other encoder models) process an input sequence into a encoding sequence as shown in the figure below, where self-attention heads encode the contextual words' meaning into each token representation.
These encodings are the foundation of many NLP tasks, and it's common (in BERT sequence classification) to then classify the CLS encoding into the desired classes (e.g. Biased, Neutral).
The CLS token (pooler_output) is a built in pooling mechanism, but you can also use your own pooling mechanism (e.g. averaging all the representations for a mean-pooled representation).
bert-base-uncasedhas 768 output features (for each token) and we can pass the CLS token into a (768 -> 1) dense layer.This output logit of that classification head is activated (typically with a sigmoid or softmax function), for a probability that falls between 0-1.
A threshold is sometimes applied to the output (e.g. probability > 0.5 is "Biased").
Metrics:
When evaluating models' performance at binary classification, you should try to understand the way positive (biased), negative (neutral) fall into the categories: correct (true) predictions, and incorrect (false) predictions.
Your individual requirements will guide your interpretation (e.g. maybe you REALLY want to avoid false positives).
Confusion Matrix: Used to visualize the levels of correct and incorrect classifications made, the goal
Precision:
Recall:
F1 Score:
Last updated