Education Technology

Higher Education

Coursera Unveils AI Function, CourseMatch, to Pair Universities with Existing Online Classes Amid Lockdowns

By Henry Kronk
April 15, 2020

On April 15, the online learning platform Coursera debuted a new machine learning tool called CourseMatch. It automatically pairs university faculties and educators with existing Coursera content to make the transition to online learning more efficient for institutions facing extended lockdowns.

Throughout 2020, the coronavirus pandemic has continued to close schools around the world. Institutions of higher ed have been scrambling to maintain instructional continuity and bring face-to-face learning online as quickly and seamlessly as possible. Many have sought to streamline this by directing their students to existing online course material available on platforms like Coursera. But the transition hasn’t always been seamless, and it isn’t always readily apparent whether an existing course will pair with one of Coursera’s.

Coursera Unveils CourseMatch to Help Universities Discover Relevant Online Learning Resources

“When we launched our Coronavirus Response Initiative on March 12, we wanted to support universities as they made the rapid transition to online learning,” said Coursera VP of Data Science Emily Glassberg Sands. “Since then, more than 2,600 colleges and universities around the world have launched Coursera for Campus programs. In addition to access, we noticed an urgent need among many institutions to identify courses in a quick and scalable way on Coursera that most closely match on-campus course offerings.

“Manual curation is too slow when it’s to be done across thousands of universities and millions of on-campus courses, especially when faculty and staff are already stretched thin. To address this need, the Data Science team at Coursera developed a natural language processing solution to automate the matching and minimize the need for human curation. We’re excited to launch CourseMatch to the world today.”

Henry Kronk: Tell me more about the logic behind CourseMatch. What information does it draw from to match an institution’s courses with those available on Coursera?

Emily Glassberg Sands: CourseMatch uses techniques from natural language processing and is powered by pre-trained word embeddings to find the courses on Coursera that are most semantically similar to each course in an on-campus catalog. For on-campus courses, the algorithms consider the course title and description from the publicly available catalog alongside more detailed syllabi and learning objectives where provided. For courses on Coursera, the algorithms consider the full text corpus – from course title and description to the lecture transcripts, assignments, and assessments.

Developing New Machine Learning and Intelligent Tools to Encourage Learning

HK: Many online learning platforms use machine learning to match learners with courses or tutors. Is this something that Coursera had considered in the past? Do you see advantages or uses to CourseMatch outside of the current pandemic? Is there potential to extend this matching technology, perhaps, to individual learners or corporate partners?

EGS: Coursera already uses machine learning to power content discovery — from search ranking to personalized recommendations across our full catalog, from courses to degrees. We also use deep learning to provide learners in-course help  with a suite of behavioral and pedagogical nudges to support each learner through the learning journey. We are continually developing new machine learning applications to meet the needs of learners, educators, and employers on the platform.

HK: Do you envision matching learners to courses based on their platform-generated data?

EGS: As this is the first iteration of CourseMatch, it will evolve based on institution and learner feedback. The main goal of this tool is to support the transition from on-campus to online learning, which does not rely on an individual learner’s platform-generated data. However, as institutions expand the blended learning model, I could certainly imagine a greater need for curriculum mapping tools based on individual majors and, at the most granular level, individual learner paths.

Image courtesy of Coursera.

One Comment

  1. I am a former summit learning teacher in Holyoke, MA. I can tell you, unequivocally, that the entire platform stinks. It is not even a curriculum, it is a hodgepodge resources lifted from Khan Academy, youtube, Engage NY, IXL lessons, scanned textbook pages, and other unrelated sources. These materials are often not aligned to common core standards, they are often of poor quality, they include numerous broken links. Students are expected to independently take notes as they work, but no consideration has been given to the lexile levels of readings so the material is often completely inaccessible to students. The math curriculum is devoid of any meaningful direct instruction. Many students disengage within a couple of weeks and spend most of their time browsing the internet or gaming instead of learning. As they fall behind, they see their home screen turn more and more red, causing greater frustration and discouragement. Students become so screen addicted that they rebel any time a teacher attempts to give them direct instruction. Worse yet, the necessity of teacher training in the platform’s usage necessitates the hiring of several consultants and coaches, many of whom explicitly state that their primary objective is to prove the platform viable so that it may grow to more school districts. Ultimately, school administrators are pressured to increase scores of online tests (many of which students attempt literally dozens of times over), so they pressure teachers to take tests with their students to ensure a passing grade. Essentially, schools are falsifying data to ensure Summit’s growth. Given that Summit pitches its product as a turnaround model for struggling urban schools, its practices are essentially exploitative.

Leave a Reply