Finding the Pattern: How Machine Learning is Helping Researchers Navigate STEM Education

Breakthroughs in research often begin by building on the work that came before. However, as professionals publish an ever-growing body of research articles every year, keeping up with the output of scientific literature has become a major hurdle. With this growing number of articles, it has become increasingly difficult to identify patterns in data sets and compare work to previous studies.  For Emily Bolger, a recent Ph.D. graduate from Michigan State University's Department of Computational Mathematics, Science, and Engineering (CMSE), that challenge has become the core focus of her dissertation.

Bolger's research explores how methods in machine learning can help researchers organize and navigate thousands of STEM education articles, making it easier to identify common trends and themes across an endless library of scientific literature. Her research builds on a 2011 study that examined how STEM educators changed their classroom approaches and how those changes improved student learning. A decade later, the research had grown with over 9,000 relevant studies covering the topic.  Before analysis on these papers could begin, thousands of titles had to be screened manually by researchers to determine which studies were relevant. Those were then narrowed down to 200 papers that required in-depth analysis. That is when Bolger decided to see how a branch of artificial intelligence (AI) called natural language processing could aid in the data sorting process, freeing up the time of highly skilled researchers to conduct detailed analysis on the content of the articles.  

"My dissertation project was trying to explore the integration of machine learning methods into the literature review process,” Bolger explained. “Could I effectively use things from natural language processing to pull out themes for papers?"  

Natural language processing allows computers to understand and analyze written language. Bolger used a machine learning technique known as an embedding to convert each article into a mathematical representation. This allowed the algorithms to group similar papers together to give researchers and STEM educators a base framework to identify themes and locate relevant articles more efficiently.

"My work pushed the bounds of what it means to include a human in the loop with AI," Bolger said. "Using the expertise of our qualitative researchers really made the model -building and output stronger."

That outlook on her research was paramount to her dissertation as a whole. While machine learning played a vital role in organizing large amounts of data, researchers remained at the heart of the operation in terms of interpreting the results, evaluating themes, and implementing these findings into STEM courses and educational practices.  

This process also gave researchers another avenue to identify patterns in how STEM is being taught in higher education. Bolger found that her machine learning models can give STEM education professionals a more efficient way to find relevant studies that could influence classroom practices.  

"I don't think it's perfect, and I don't think we should fully trust it,” she acknowledged. “But I do think we can leverage the benefits of being able to synthesize large amounts of text very quickly." 

While machine learning tools streamlined her workflows, many models came with overwhelming computational demands that a standard laptop could not accommodate. To counteract this dilemma, Bolger turned to the resources and facilities provided by the MSU Institute for Cyber-Enabled Research (ICER). ICER's Data Machine allowed her to process her embeddings and test her clusters of data at a much more efficient rate. 

"The Data Machine meant that I wasn't being held up by any sort of computation time," she said.

For Bolger, ICER's impact on her research extended far beyond processing power. The user-friendly tools at her disposal, such as skipping the command line and using the OnDemand graphical user interface, made advanced computing more accessible, which in turn allowed her to spend more time on her research rather than the technical hurdles that can come with high-performance computing systems.  

"I think it really enables researchers who may not have a good understanding of internal terminal structure to use these complex models," she noted.  

Bolger was first introduced to ICER when she participated in the ACRES Research Experience for Undergraduates. Not only did this experience spark her interest in interdisciplinary research, but it also created the early framework for the computational skills she would later develop in graduate school.  As her research required higher standards in computational power, ICER remained a trusted resource, providing her with the computing resources and support needed to elevate her dissertation. Throughout this process, Bolger discovered that ICER not only provided her with computing systems, but also peers and mentors.  

"There's a lot of resources—not just computing resources, but people's resources—that you can leverage,” she explained. “It gives you exposure to what a research project should look like. Even though it seems scary, I think just trying it and using the resources that are available to you is the advice that I would give."

As she prepares to begin a position as a data science preceptor at the University of Chicago, Bolger plans to continue investigating how methods in machine learning can improve STEM education for the next generation. As scientific literature continues to grow, Bolger hopes that researchers will embrace artificial intelligence as a tool for furthering scientific understanding, rather than a replacement for human knowledge.