Using machine learning to discover university writing contexts: Evidence from over 600,000 scraped syllabi – UROP Spring Symposium 2023

Using machine learning to discover university writing contexts: Evidence from over 600,000 scraped syllabi

Alessandro Lorenzo

Alessandro Lorenzo photo

Pronouns:

Research Mentor(s): Jason Godfrey
Research Mentor School/College/Department: Joint Program in English and Education / LSA
Program: UROP
Session: Session 3 (11:00am – 11:50am)
Authors: Alessandro Lorenzo, Jason Godfrey

Abstract

Writing is a significant component in higher education; however, currently, we’re not exactly sure where or how often writing occurs within collegiate institutions. To address this issue, we leverage data made available through Texas’s 2010 House Bill 2504, requiring all public institutions to post their course syllabi online. Thus, using Python, we have web scraped 345,289 syllabi from public universities in Texas. Our project is focused on analyzing and annotating data within the syllabi derived from five different public postsecondary institutions. Taking a text as data approach, our goal is to determine whether these syllabi, and by extension, courses, have a substantial writing component or not; however, this is more so a test case approach as it can be used to identify any significant component within these documents. Given the enormous size of our data, we use machine learning and natural language processing to scale our annotations. In an effort to train our machine learning model to annotate and identify if a syllabus contains a substantial writing component, we use a codebook that lets human annotators attach labels to syllabi and have had four humans annotate a random sample of 500 syllabi from each school in our dataset. Our method consists of an 80/20 training test split for our model so it can learn from supplied data and make data-driven predictions based on certain criteria. Our project is still evolving through continued research, annotation, and modeling. Nevertheless, preliminary analysis suggests two interesting findings. One being that 18-21% of courses contain a substantial writing component which is a number that can vary greatly by discipline, and the other being that well-resourced schools assign more writing and assign it more often than other more under-developed schools.

Engineering

lsa logoum logo