csc522-character-biases

CSC522 Project - Gender and Race biases

This project Identifies Gender and Race biases in Audience Perception. The data for the project is collected from OpenPsychometrics.org, enriched with Betafaceapi.com and is present inside the data directory.

Abstract

People may have implicit biases on their perception and classification of character traits based on attributes such as race and gender. Furthermore, the portrayal of women and characters of color is a subject of much controversy and discussion in recent years. In this study, we propose examining and measuring such racial and gender biases in audience perception of characters by using Automated Learning and Data Analysis techniques, which may also allow us investigate the nature of those stereotypes. Data obtained from OpenPsychometrics.org contains audience ratings of 1600 characters along with hundreds of pairs of bipolar personality adjectives, which are crowd-sourced through online surveys.

Using a semi-supervised approach of Betafaceapi and manual evaluation and adjustments, OpenPsychometrics data was enriched with race and gender information to have reliable target information. We engineered two models: one that predicts character's gender, and another that predicts race, using the adjective scores. Cluster analysis to visualize the presence of stereotypes was also performed.

The expectation is to support that we can predict character race and gender based only on personality traits, suggesting that stereotypes are present in the portrayal or audience perception of characters.

Notebooks

Some of the notebooks present in this repository:

• Scraper - Used to fetch data from OpenPsychometrics, and save in the data directory.

• PhotoLabels - Scraper that uses Betafaceapi to add labels to the images downloaded from OpenPsychometrics

• ImageLabeling - Tool used to hand label gender/race on images.

• EDA / EDA-OpenData - Exploration and Data Analysis documents, where we do parameter cleaning and joining them from both sources, and export into a single CSV.

• Cluster-Analysis - Used to analyse possible clusters within the dataset, using unsupervised learning methods.

• Model-Exploration-* - Gets exported data from EDA, executes and compares various classifiers with the hand labeled data with supervised learning methods. The most updated one is Model-Exploration-OpenData.