Race and gender biases have been a long-standing issue in paper textbooks, but prior research showed that online literacy development materials still carry significant biases and misrepresentations, potentially exacerbating the problem at a much larger scale. Surfacing these biases and stereotypes thus becomes crucial in flagging problematic content and ensuring the integrity of these learning materials. This study applies NLP techniques to nearly 2000 news articles used as literacy development materials by a popular K-12 digital learning platform to answer two following research questions: 1) which race and gender groups are more associated with which topics, and 2) how similar are the contexts in which different race and gender groups appear? Results from topic modeling show that different gender and race groups are associated with widely different set of topics: men are associated more with “crime and law enforcement”, “politics”, “healthcare/medicine”, and “war”, whereas women are associated more with “education”, and “international issues”; white people with “politics” and “government/national issues”; black people with “crime and law enforcement”; Asian and Latinx people with “minority” and “immigration”. In addition, results from word embedding show that different race groups don’t often appear in similar contexts, indicating potentially narrow and stereotypical portrayals of minorities, consistent with prior research. Overall, this study contributes to a more nuanced understanding of race and gender misrepresentations in digital literacy development materials and highlights the importance of culturally responsive education in the digital age and the need for a more inclusive and just education for future generations.
© Stanford University, Stanford, California 94305.
