By Rachsuda Jiamthapthaksin 10092009 RESEARCH CHALLENGES IN RECOMMENDER
By Rachsuda Jiamthapthaksin 10/09/2009 RESEARCH CHALLENGES IN RECOMMENDER SYSTEMS / SURVEY OF THE NETFLIX CONTEST Edited by Christoph F. Eick 1
Recommender Systems (RSs) Goal: To help users to find items that they likely appreciate (and buy/lease) from huge catalogues. 2
The recommendation problem Let ○ C be the set of all users, and ○ S be the set of all possible items that can be recommended. ○ u be a utility function that measures the usefulness of item s to user c, u: C S R For c C, find s’ S that maximizes the user’s utility: c C, s’c = argmaxs S u(c, s) (1). 3
Netflix Recommender System Scenario : = unknown Remark: Typically, a lot of symbols 4
Survey of the Netflix Contest Netflix Prize competition offers a grand prize of US $1 M for an algorithm that’s 10% more accurate than “Cinematch” Netflix uses to predict customers’ movie preferences. The best score will win a $50 K Progress Prize. 5
The Basic Structure of the Contest Provide 100 million ratings that 480 K anonymous customers had given to 17 K movies. Withhold 3 M of the most recent ratings and ask the contestants to predict them. Assess each contestant’s 3 M predictions by comparing predictions with actual ratings. Evaluation metric: the Root-Mean Squared Error 6
Netflix Dataset (1) The data were collected between October, 1998 and December, 2005 and reflect the distribution of all ratings received during this period. The ratings are on a scale from 1 to 5 (integral) stars. The date of each rating and the title and year of release for each movie id are also provided. 7
Netflix Dataset (2) training_set. tar (2 GB) movie_titles. txt (575 KB) qualifying. txt (51, 224 KB) probe. txt (10, 530 KB) rmse. pl (1 KB) 8
- Slides: 8