Recommender systems have become an important part of modern internet life. When using online stores like Amazon it’s clear how useful recommender systems can be. The world’s most popular websites from iTunes to Netflix, Last.FM to TripAdvisor use recommender systems to help ordinary users cope with what would otherwise be an overwhelming range of choices.
UCD has been granted patents by the US and Chinese patent authorities for newly developed recommender systems which emerged from research undertaken by Professor Barry Smyth, Director of CLARITY (The Centre for Sensor Web Technologies) and Digital Professor of Computer Science at the UCD School of Computer Science and Informatics.
These patents were filed by NovaUCD, the Innovation and Technology Transfer Centre on behalf of UCD and Professor Smyth. The patents were co-funded by NovaUCD and Enterprise Ireland under the Higher Education Patent Fund.
“The patents describe a new technique for solving the so-called diversity problem that exists with conventional recommender systems,” says Professor Smyth. “The new technique will help existing recommender systems to produce recommendations that are both relevant to users and different from each other. This delivers users a broader set of relevant suggestions, and provides online services/stores with a greater chance of success when it comes to presenting the user with suggestions they will like. Improving the quality of recommendations in this way will lead to more satisfied customers and improved online sales for retailers.”
For a given user, conventional recommender systems generate suggestions by picking items (e.g. movies, books, or music) that are like items that other similar users have enjoyed in the past; for example, we are all familiar with Amazon’s suggestions about what other similar users have recently purchased. “While variations on this basic theme work well for recommender systems in practice, one limitation is a tendency to produce sets of recommendations that are similar to each other,” explains Professor Smyth.
Among the top recommendations that a Harry Potter fan might receive could be “Harry Potter and the Deathly Hallows” (Book 7), “Harry Potter and the Half Blood Prince” (Book 6) and perhaps the latest DVD release for “Harry Potter and the Order of the Phoenix” (Book 5).
According to Professor Smyth, these recommendations are all very similar to each other and if the user is not interested in Harry Potter, at this particular point in time, then the recommender system will have wasted valuable screen-space by making these near-duplicate suggestions.
“A more effective approach involves selecting one Harry Potter suggestion and then complement this with other recommendations that are also relevant to the user but that satisfy different tastes and preferences that they might have. The new technology that has just been granted patents in China and the US describes how to do this, and in a way that is compatible with existing recommendation techniques. It can be readily applied in the marketplace.”
Recommender systems has benefited from significant innovation over the past 15 years. It is among the most active research fields within computer science. Current recommender systems technology has graduated from a number of leading international academic research institutes to make a significant impact on the commercial marketplace. In 2006, Netflix (the leading US online movie rental company) announced a $1 million recommender systems prize as part of a 5-year competition to develop the next generation of recommender system. This competition continues to attract new entrants as it enters its third year; see www.netflix.com.
According to Prof. Smyth, Irish researchers are at the forefront on this research field. “Research groups in University College Dublin and University College Cork are ranked among the leaders in recommender systems research,” says Prof. Smyth. “This has led to significant innovation in the marketplace. Over the past 10 years we have seen Irish companies such as ChangingWorlds playing a leading role in successfully bringing this technology millions of consumers.”
ENDS
April 23 2009
For further information contact Micéal Whelan, NovaUCD, e: miceal.whelan@ucd.ie, t: (01) 716 3712.
