Characterizing gender stereotypes in popular fiction: A machine learning approach

Abstract

Gender representation portrayed in popular mass media is known to reflect and reinforce societal gender stereotypes. This research uses two methods of natural language processing–Word2Vec and bidirectional encoder representations from transformers (BERT) model–to analyze gender representation in popular fiction and quantify gender bias with gender bias score. Word2Vec, which represents the words in vectorized format, can capture implicit human gender bias with the geometry relationship between word vectors. BERT, a newer pre-trained deep learning model, is specialized in understanding words in the larger context it appears in. The research will compare the results obtained from Word2Vec and BERT. With book check out records from the Seattle Public Library checkout dataset–an ongoing open source dataset from the public library system of Seattle, WA–the research aims to identify evolutionary trends of gender bias in popular fiction and analyze consumer preferences regarding gender representation.