tritical's Netflix Page


Current RMSE: 0.8822
Current Rank: 21
email
Netflix Leaderboard

About Me

My name is Kevin Stone, and I'm 24 years old. I'm currently a graduate student (working on a master's degree) in the Electrical and Computer Engineering department at the University of Missouri - Columbia. I'm working on the Netflix Prize as part of CS 8750 (Artificial Intelligence II).

Techniques Used

My 0.8822 submission is a blend of 32 prediction sets generated using the following methods:

  1. Movie-movie collaborative filtering using pearson correlation on user-based zscores and on double centered data (PRMSE = 0.9236-0.9249)
  2. Cluster based user-user collaborative filtering on double centered data (PRMSE = 0.9181)
  3. Regularized matrix factorization with biases on raw data (gradient descent and alternating least squares training) (PRMSE = 0.9058-0.9166)
  4. Paterek's NSVD1/NSVD2 asymmetric factor models on double centered data (PRMSE = 0.9375-0.9613)
  5. Regularized non-negative matrix factorization on raw data (PRMSE = 0.9164-0.9324)
  6. Movie-movie collaborative filtering post-processing on residuals of 2, 3, 4, and 5 (PRMSE = 0.9004-0.9215)
PRMSE is the probe set RMSE when training without the probe included. "Double centered" means removal of movie and user means (i.e. first two global effects).


Last updated: 4/1/08

Avisynth Filters Page