Worcester Polytechnic Institute (WPI) PhD student Zach Pardos placed second among student teams and fourth place overall out of more than 600 teams, in the 2010 Knowledge Discovery and Data-mining (KDD) Cup – a two-month-long, high-profile annual data mining competition run by the Association of Computing Machinery (ACM). His performance at the KDD Cup allowed him the chance to present his research findings this week in Washington, D.C.
Data mining is the process of analyzing data from different perspectives and summarizing it into useful information, and, for his achievement at the KDD Cup, Pardos received $3,000 in prize money and also travel funds to attend the July 25-28 Special Interest Group on Knowledge Discovery and Data Mining (SIGKDD) Conference in the nation's capital. At the conference, Pardos, a native of Denver, Colo., had several opportunities to present his research, which is under consideration to be published in an upcoming issue of the Journal of Machine Learning Research.
Pardos' paper, which describes one of the methods employed in the competition, was based on his work that won second place in the Science Division at WPI’s 2010 Graduate Research Achievement Day. Funded in part by the National Science Foundation and the U.S. Department of Education, his research findings allow for better models of student learning, which has significant practical implications. With a better model of learning, existing educational software can be improved in a variety of ways.
The ultimate goal, says the 29-year-old Pardos, is for his research to culminate in the United States being more competitive with the rest of the world, with respect to math and science education, teaching methods, and learning, "but we need the technology to do that," he explains. "Let's improve how every student learns, not just how the average student learns. To do that, you have to adapt to an individual's learning style, like a one-on-one tutor does."
The current state of the art in the field has been "Knowledge Tracing," a technique developed in 1995 by Albert T. Corbett and John R. Anderson, according to Pardos' advisor, WPI Computer Science Associate Professor Neil T. Heffernan. "It is used by millions of students across America in many different pieces of educational software to track student knowledge," he says. "Zach's new method, which served as the foundation for his KDD solution, is a model that tracks learning rates, guessing rates and other characteristics of the individual user. The presentation of Zach's paper at this summer's User Modeling conference in Hawaii shows how this new model definitively outperforms the 'Knowledge Tracing' approach."
At WPI, Pardos works with Heffernan to develop Intelligent Tutoring Systems. A team of scientists, led by Heffernan, researched and developed a program called ASSISTment, which is a powerful intelligent mathematics tutoring system developed at WPI; it’s helping students learn and schools track students' progress. Funded by more than $9 million in federal research funds, the system helps students learn mathematics by presenting them with problems and then offering carefully structured assistance if they struggle. By recording students' attempts to answer questions and the help they request, the system also assesses which concepts they have mastered and which they still need to work on. It has been adopted for use throughout grade 7 and 8 algebra classes in the Worcester Public School System as well as in schools in Fitchburg, Leicester, and Shrewsbury.
Referring to Pardos' accomplishments at the KDD Cup, Heffernan said: "This is a very prominent achievement. Many times engineering competitions lead to practical solutions but not always good science. Not only did Zach do well, but he has also done important scientific work that will lead to several publications comparing multiple different modeling approaches."
Pardos earned his bachelor and master's degrees at WPI. He says that his studies at the university have helped prepare him for his doctoral research. "I'm happy that the skills that I've learned and applied in the WPI community are also formidable on the world stage," he said.
Some of the best data mining teams in the world compete each year at the KDD Cup to solve a practical data mining problem. This year's challenge was to focus on how generally or narrowly, and how quickly or slowly, do students learn. Participants were asked to predict student performance on mathematical problems from logs of student interaction with Intelligent Tutoring Systems, which are data mining software tools that analyze data from different dimensions or angles, categorize it, and summarize the relationships identified. The annual ACM SIGKDD conference in Washington, D.C., is the premier international forum for data mining researchers and practitioners from academia, industry, and government to share their ideas, research results and experiences. This year's conference features keynote presentations, oral paper presentations, poster sessions, workshops, tutorials, panels, exhibits, demonstrations, and the KDD Cup competition.
The New York Times’ “Bits” blog covered this week's conference.