Knewton Adaptive Learning
On March 18 we were lucky to host Chaitu Ekanadham, a data scientist from Knewton. Knewton is an education technology company that uses adaptive learning techniques. It uses data about past learning to aid students in future learning experiences.
For example, in traditional “text book” learning, students move through the content in a linear fashion, one chapter and then the next. While this is a tried and true method of teaching, it isn’t the best solution for every student in every situation. Some students may need only a review of some particular content, while others may need to spend a great deal of time on the same content. For that reason, it makes sense to direct students to content dynamically--show them only the content that would be most helpful for them at that time.
When Knewton begins working with a publisher, they start by organizing the material in the textbook into a Knewton knowledge graph. This is used to represent the ways that content is related to each other conceptually. In this way students’ progression can be evaluated automatically. This gets the right content to the right student at the right time.
So how does one even begin to apply models to this sort of information? Knewton primarily utilizes the data to create models describing the learners, and models describing the educational content.
The model of the learner is individualized to each student. After a student has completed some amount of graded content, the system has some idea of their capability. Knewton can then model the likelihood that a student will correctly answer some future question. Quiz questions can then be presented that are neither too hard nor too easy.
When recommending new content for a student, the system avoids any content that is predicted to have a likelihood of either close to 0 (indicating the student will most likely not learn or retain this information) or 1 (which indicates that the student is familiar with the concepts and does not need any further instruction).
All material is given a score for how difficult it is based on the number of students who historically found it difficult. This value is used in the above model of likelihood a student will get it correct.
In addition, the content is judged for how it engages the students. If students generally move through the content with few breaks it is considered engaging – while content that is consistently associated with long breaks, is considered not engaging. Response time is summed up for all students to give a value for a typical response time for some given content.
How These Models are Used
Knewton uses this data in several different business purposes
1. Generate recommendations for students--what is the best content for that student to consume right now?
2. Analytics--are students on track? What do they need to do in order to pass the next assessment of their skills?
3. Content insights for the creators--what content is most effective? Are there any gaps in content?
4. Classroom dashboards for teachers--how is the class doing? Who is falling behind?