Sign In

Communications of the ACM

ACM TechNews

Improving Recommendation Systems

View as: Print Mobile App Share:
Recommendation systems

Image courtesy of Christine Daniloff

Massachusetts Institute of Technology (MIT) professor Devavrat Shah thinks that the most common approach to recommendation systems is fundamentally flawed.

Shah says recommendation systems should ask users to compare products in pairs instead of using the standard five-star scale, as both Amazon and Netflix do. He believes that combining the rankings into an overall list will offer a more accurate representation of consumers' preferences.

Shah, MIT professor Vivek Farias, and students Ammar Ammar and Srikanth Jagabathula have demonstrated algorithms that put the theory into practice and they have created a Web site that uses the algorithms to help large groups make collective decisions. The MIT algorithm predicted car buyers' preferences with 20 percent greater accuracy than existing algorithms. The algorithm reduces the number of possible orderings, which would be more than 3 million for a list of just 10 items, by throwing out the outliers and selecting subsets of data.

The algorithm also uses an items rank in each of the orderings, combined with the probability of that ordering, to give the item an overall score, which is used to determine the final ordering.

"They've really, substantially enlarged the class of choice models that you can work with," says University of Southern California professor Paat Rusmevichientong. "Before, people never thought that it was possible to have rich, complex choice models like this."

From MIT News
View Full Article

Abstracts Copyright © 2011 Information Inc. External Link, Bethesda, Maryland, USA 


No entries found