Question
Note to players: The answer to this tossup is pretty general. In one technique used for this task, Alternating Least Squares is usually preferred to Stochastic Gradient Descent for optimization due to its parallelizability. A Simon Funk blog post popularized a class of modified SVD algorithms for this general task, such as timeSVD and SVD++. Hybrid methods might combine algorithms for this task with knowledge-based approaches to mitigate the “cold start” problem. Latent factors in user data are found with matrix factorization in an example of the collaborative filtering method for this task. The GroupLens lab maintains a popular dataset for evaluating algorithms for this task, collected from their website (*) MovieLens. The team “Bellkor’s Pragmatic Chaos” won a million dollar prize for improving an instance of this task by 10 percent in a contest run by Netflix. For 10 points, the huge success of TikTok has largely been attributed to them perfecting algorithms for what task? ■END■
ANSWER: recommendation [accept word forms; accept collaborative filtering before mention; accept descriptive answers of predicting what users like; accept descriptive answers of choosing which items to show to a user; accept more specific answers like recommending videos; prompt on “personalization”; prompt on “ranking” or “filtering” with “as applied to what task?”; prompt on “matrix factorization”, “matrix decomposition”, “singular value decomposition”, or “matrix completion” with “as used for what general task?”; prompt on “learning” with “learning what?”]
<JX and AW>
= Average correct buzz position
| Conv. % | Power % | Average Buzz |
|---|
| 100% | 0% | 123.00 |
Back to tossups