Think of it. Educational technology has become a huge market. A generation of students of all ages is learning geometry with computer tutorials, or foreign languages with CD-ROMs, or biology with the help of reference software. These days, more and more people learn by clicking. Many of the educational programs record response times to their tutorial questions and accuracy in responses. Can that information provide insights into how people learn and how computers teach us?

It's those questions that have led to a new field of study: educational data mining. Last summer, the First International Conference on Educational Data Mining took place in Montreal, Canada. At the conference, geared for researchers from computer science, education, psychology, and psychometrics, the award for best paper went to "A Response Time Model for Bottom-Out Hints as Worked Examples." The paper was authored by an interdisciplinary trio from Carnegie Mellon—Benjamin Shih, doctoral student in machine learning; Kenneth Koedinger, professor of cognitive psychology; and Richard Scheines, head of the Department of Philosophy.

The research examined whether learning takes place when a user incorrectly answers a tutorial question and is then given hints by the software. The moderate response time in conjunction with the accuracy data indicate that learning does take place, even when software essentially gives away the answer with a "bottom-out hint" after a user makes several incorrect responses to the same question. Had the incorrect answers, prompting easier and easier hints, occurred with little time elapsing, it would have indicated that the user was simply trying to get the software to reveal the answer.

Koedinger, pleased with his team's recognition, says he was surprised to learn that it may be okay when students miss a question on the first try. Think of that.

—BRADLEY A. PORTER (HS'08)