Content Analysis of Textbooks via Natural Language Processing

Purpose: Advanced methods from the field of data science have the potential to shed fresh light on basic concerns in the field of educational research. Natural language processing tools, such as lexicons, word embeddings, and topic models, are used in 15 United States history textbooks that were extensively used in Texas between the years 2015 and 2017. Material and Methods: This study aims to analyze these textbooks for their portrayal of historically oppressed populations. Latinx individuals are rarely mentioned, but renowned white men are almost always mentioned. People of African descent are characterized as behaving in ways that imply helplessness and lack of control, according to lexical methods. Women are most often addressed in the contexts of the home and the workplace, according to the word embeddings. Issues of a political rather than a social nature are highlighted by subject modeling. Findings: We also found that textbooks with a smaller representation of women and people of African heritage are more often purchased by conservative nations. Implications to Theory, Practice and Policy: Our computational toolkit has a rich history of textbook analysis and has recently been distributed as part of our efforts to support new fields of study.