In a typical classroom there may be several students who work at a higher grade level, and several who work at a lower one. Student don’t learn at the same pace, so their individual needs are often neglected because teachers can’t create personalized lesson plans for every student and work with them directly on what they’re struggling with, or challenge them. Even if the teacher did have time to create personalized lessons plans, the data available to them is usually only from a student’s latest test, or annual standardized tests. To combat this problem, several education startups are using big data to facilitate adaptive learning.
There are hundreds of thousands of data points for each student over the course of their academic career, data points which could be leveraged to deliver customized education. This type of learning, which adjusts based on student data, is called adaptive learning. While some types of adaptive learning will ask a set of questions to customize the course material, true adaptive learning will adjust every question based on a student’s previous answer. Think of it as Pandora for learning.
Knewton is a startup founded by Jose Ferreira, a former executive at test preparation materials company Kaplan, and it’s aiming to use big data to provide insights into how people learn. The Knewton platform gives schools, publishers, and developers the ability to provide adaptive learning for any student. In October 2011 they closed a $33 million Series D round led by Founders Fund and textbook publisher Pearson. Knewton is working on having educational content tagged so it can be placed into a “Knowledge Graph.” This system determines what concepts need to be learned before a student can move on to others, and how they all fit together.
“We’re getting publishers and content providers to tag their content at a very granular level. When it is tagged we can break it all down and provide it to the user when they need it, which is a continuous process,” said Knewton COO David Liu in an interview. The company recently parterned with Pearson to tag every textbook under their imprint work with the Knewton Knowledge Graph.
The technology seems straightforward for math and sciences, but what about classes which don’t have a clear-cut right or wrong like history or dramatic arts? “You have to establish a rubric around what constitutes proficiency and every concept within the subjects. We’re doing that now with history and other liberal arts and once we establish standards and levels of proficiency we can adapt,” Liu said. There are certain highly subjective types of content like poetry or ballet which Liu agrees can’t be effectively measured, but for the vast majority of education scenarios that exist they’re able to make them adaptive.
The second tool they’re building allows them to do data mining and take various inputs, like test question results, activity on the system, what links students clicked, etc. to make a prediction of the next best piece of content for a student to learn. Typical systems will take one point of measurement and use human adjustment of course content, but Knewton does it on the fly after every question.
The technology seems to be working. After a pilot project at Arizona State University with 5,000 remedial math students, pass rates improved from 66 percent to 75 percent, with half the class finishing four weeks early. Liu states that the school is grouping students based on their performance in previous courses.
“The professors are much better prepared for a single class so that they can give much more individualized instruction,” Lui said. “The practical effectiveness of this means that teachers are now able to use their time more efficiently to hone in on the things that are most troublesome or useful for different groups of students. You’re not teaching to the mean or bottom quartile.”