A personalized touch can make all the difference.
When you log into Amazon, Netflix, or Facebook, one of the first things you see are recommendations for products, shows, or friends you may know—all based on things you have already bought, watched, or liked.
Recommender systems have eliminated the time-consuming effort of understanding and anticipating what exactly users want, sometimes before they know they want it. Using vast collections of detailed data points, data scientists can create a trail of digital breadcrumbs, which follows Internet users as each sale, search, and interaction becomes part of an algorithm for new suggestions. These platforms can predict and encourage your next shopping sprees, binges, and bucket lists.
In the private sector, companies have long been using technology-enhanced learning in order to proactively suggest and anticipate their consumers’ choices. But how can these mechanisms be most effectively applied to academia?
“We have to learn something new every day,” says Dr. Konstantin Bauman, assistant professor in the Management Information Systems (MIS) Department at the Fox School of Business. “With the traditional path, we go to university, take some courses, meet with the instructor once a week, and go to lectures. But today, there are many different types of tools and materials available online that are able to educate large groups in a personalized and direct way.”
Educational companies like Coursera, Lynda.com and the Khan Academy already use recommender systems to suggest courses their users may like, based on their history. Bauman, however, wanted to know whether personalized e-learning can help students struggling to comprehend a particular subject.
“Education is one important application of recommender systems for society to target materials for specific learning paths,” says Bauman.
Bauman, along with co-author Alexander Tuzhilin of New York University’s Stern School of Business, examined the personalized e-learning systems approach with the help of a tuition-free online university. By using curricula from 42 classes and test results from 910 students over three semesters, the researchers created a system to pinpoint—and address—specific areas within a student’s comprehension that needed improvement.
The team reported their findings in the paper, “Recommending Remedial Learning Materials to Students by Filling Their Knowledge Gaps,” which was published in MIS Quarterly in 2018.
In this real-life experiment throughout the 2014-2015 academic year, Bauman’s team identified where students’ knowledge waned and provided materials to supplement these gaps. “Instead of revising the entire lesson, we provided catered lessons to fill those gaps,” says Bauman.
The students had diverse backgrounds, from both the United States and developing countries, and studied in a variety of programs, from business to computer science to art history. The researchers split the students into three groups: a control group that received no recommendations, a group that received non-personalized recommendations, and a group that received recommendations tailored to the individual student.
Bauman and his team created taxonomies that mapped all the topics covered within a specific course, built a library of remedial learning materials, and matched test questions with course topics. After analyzing test scores, the researchers identified the students’ weaknesses. The students in the non-personalized group received generic recommendations for the course, while students in the personalized group received remedial materials for specific topics that were identified via testing.
“First, we showed that most of the students who received our recommendations found them relevant and helpful,” says Bauman. Second, the “average” students, who received a test score between 70 and 90 in previously taken courses, were most affected by personalized recommendations. “These students improved their performance on the final exams significantly more, in comparison to their prior performance before they received personalized recommendations than the students from the control group.” For this subset of students, the personalized group received an average grade of 83.22 in their final exams, while the control group scored an average of 79.39.
The study received limited interactions with students who were classified as “falling behind” (those whose previous grade averages were below 70) as only six students who received personalized recommendations actually clicked on the materials. Similarly, students who were “excellent” (with average grades above 90) were less likely to need remedial lessons.
Bauman found that, by determining specific materials needed to supplement their understanding, students saved time and energy in preparation for their exams.
“Learning systems have the capability of picking up patterns and behaviors that can clearly predict necessary methods that are worthwhile and timely,” says Bauman. For students and professors, time that may be used to teach a specific lesson can be accomplished through recommender systems, saving more time for interactions that encourage new ideas and understandings.
One thing is for sure—when it’s time to come back for more, a new suggestion will be waiting.
This story was originally published in On the Verge, the Fox School’s flagship research magazine. For more stories, visit www.fox.temple.edu/ontheverge.