by Angela Guess
Sarah Kessler of Mashable recently explained why Kaggle’s model of hosting contests for data scientists is working so well. She writes, “In 2006, Netflix released a set of anonymous user data and a challenge: Anyone who could improve its movie recommendation algorithms by at least 10% would receive a $10 million prize. It took three years, but a team (two teams, actually) eventually did meet the goal — something Netflix’s own data scientists hadn’t been able to do on their own.”
Kessler quotes Kaggle CEO Anthony Goldbloom, who explains, “When you’re working in the context of a competition, you try a few things and get to the top of a leaderboard and you’re pretty happy with yourself. Then someone else comes along and displaces you, so you need to try something else. This leap-frogging generally continues until nobody can get any better — it’s like squeezing a lemon until it’s dry.”
Kessler continues, ” It’s not just Netflix that relies on predictive models — banks, insurance companies, test providers and numerous other industries do too. Goldbloom took the Netflix Prize model and made it viable for all types of companies. Kaggle hosts contests for data scientists. Companies who want problems solved post them, along with relevant data sets, on the site. Anyone can submit a solution, and each competitor ranks on a leaderboard throughout the competition. So far about 30,000 people have submitted at least one model to a contest.”
Image: Courtesy Kaggle
