Knewton is applying Recommendation Systems to exercises, basically accumulating statistics on every participant behavior and route them in a directional network of concepts that they established, directions or precedence probably added through curation after extracting concepts through dbpedia for example.
As you take a test, answers to your questions are evaluated (which is probably why they now focus on very straightforward tests like GMAT or SAT) and that establishes a trace of how well and how bad you did on specific topics but also, over time, how fast you improved on those same topics.
Consequently recommendation on what lessons you have taken to made those fast improvements can be suggested to people who have a rather similar trace.
This might sound rather complex but the field of automated recommendation is now nearly 10 years old so experience has been accumulated distinguishing several techniques using profiles, item-to-item, etc and competitions like NetFlix pushed to rather high precision.
So overall and without having access to comparative results of tests against other techniques I can just say it sounds efficient to me so yes, I would say it works.
Does it work better than a book of exercises and Wikipedia or going to class, I can not say.
Building a Better Netflix for Education by Jonathan Bethune, Knewton Blog January 2011 http://www.knewton.com/blog/edte…
What is Adaptive Learning? by David Kuntz, Knewton Blog June 2010 http://www.knewton.com/blog/edte…
PS : I did not find a white paper or a patent on Knewton’s “Adaptative Learning Engine” so this is limited to my understanding of links I mentioned.
More Resources on Does Knewtons Adaptive Learning Engine work:
Read more about Does Knewtons Adaptive Learning Engine work: