Coursera’s algorithms guide and motivate to give learners a leg up — plus do what only machines can
Learning isn’t always easy. Coursera learners challenge themselves to master new concepts and demonstrate skill development by tackling lectures, readings, exams, and assignments. The majority of online learners traverse these learning journeys without in-person guidance or support, and for many the challenges prove substantial. In studying online course dropouts, we found two major categories.
The blue circle above highlights dropouts resulting from a specific pain point in the course (in this case, a programming assignment in the second week). This one assignment is responsible for about 30% of all dropouts for this course. In the classroom setting, an instructor might observe that a student is unprepared for or intimidated by an upcoming assignment and provide them with an additional study guide or a warning to allot some extra time.
Separately, the orange circle highlights dropouts who are a bit more idiosyncratic. These are learners feeling the toll of moving though long learning journeys with only self-motivation to push them forward. In-person teachers might reinvigorate their students by framing the context of their progress and highlighting especially important concepts to break up the malaise.
How can these proven coaching philosophies be scaled to the tens of millions of online learners on Coursera?
The scale created by the online learning medium creates not just a set of challenges but also a set of opportunities to solve these challenges through data. For example, historical data can quickly reveal the most important opportunities for support and coaching, and automate many of the best practices of in-person instructors.
To unlock this value for learners we’ve launched a simple in-course help feature that delivers pop-up messages to learners as they move through course material.
Hooking up the front end shown above with machine learning models on the back end, we leverage data from more than 100 million historical course enrollments to provide behavioral and pedagogical nudges that help learners succeed. Here are a few examples:
- Behavioral interventions around a course pain-point
Our algorithms predict key drop-off points and mentally prepare learners for the challenge by encouraging a growth mindset and citing the success of peers as social proof.
2. Additional context to focus learner attention
We also leverage historical learner activity to surface the most essential concepts to new learners as they work through a course.
3. Framing learner expectations
We set accurate expectations for learners at opportune moments based on the observed behavior of past learners like them, and encourage learners to plan ahead appropriately.
Each of these interventions mirrors some of the best of on-campus pedagogy and instruction strategy, and each increases the number of course items learners complete by more than 2%.
And that’s only the tip of the iceberg: Rich streams of learning data in the online context allow Coursera to go further than an in-person instructor can.
One example is providing hyper-targeted pedagogical support for learners who are struggling with a quiz or assignment. Based on the activity patterns of all prior learners, our algorithms recommend the best review material for that specific learner’s challenge. That means learners no longer have to flip through textbooks or furiously Google to find the answers they need — and they successfully pass the assignment at a 4% higher rate.
By using data to enrich teaching and learning, Coursera ensures that learners’ course experiences are continually improving. And, as each of our 33 million learners engages with the platform, his or her behavior further informs the models — and future learners reap the rewards.