Gender Biasing Algorithm

A Python-based project for identifying possible gender bias in sentences by comparing gender-related subjects and their associated objects using natural language processing and similarity analysis.

Overview

This project analyzes text to detect whether a sentence may contain gender bias. It uses:

• text preprocessing
• subject and object extraction
• gender-related word detection
• word similarity comparison using CBOW and Skip-Gram models

The goal is to estimate whether the language in a sentence leans toward a male or female association based on contextual similarity.

Project structure

• Gender_bias.py – main bias detection logic
• Data analysis code.py – data analysis and experimentation scripts
• practice.py – additional exploratory code
• text-and-id.txt – dataset used for sentence analysis
• pairs-label-training (1).txt – label data used for evaluation
• README.md – project documentation

Setup

Requirements

• Python 3.9 or later
• pip

Install dependencies

pip install -r requirements.txt
python -m spacy download en_core_web_sm
python -c "import nltk; nltk.download('punkt')"

Run the project

python "Gender_bias.py"

You can also run the analysis script:

python "Data analysis code.py"

Important note

Some scripts contain hard-coded Windows file paths to local dataset files. Before running them, update those paths to match your machine or dataset location.

Example:

file_name = 'D:\\Python\\Data sets\\...\\text-and-id.txt'

Replace this with the correct path for your environment.

Method

The project follows a basic NLP workflow:

1. Clean the text and remove punctuation.
2. Extract sentence tokens.
3. Identify subject and object phrases.
4. Detect gender-related terms.
5. Compare similarity between gender-related terms and object terms.
6. Evaluate bias based on the difference in similarity scores.

Data

The dataset contains sentence-level text and associated labels used for evaluating bias detection. The project also compares model outputs with training labels to estimate accuracy.

Limitations

This is a research-oriented prototype and is not a production-grade fairness model. It depends heavily on:

• dataset quality
• language coverage
• word embeddings
• hard-coded preprocessing rules

License

No license has been added to this repository yet.