Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 114 views

    Netflix Prize: All Together Now: A Perspective On The

    Uploaded by

    fifi
    1) The document discusses the Netflix Prize competition, which challenged participants to improve Netflix's movie recommendation algorithm (Cinematch) by at least 10% based on a dataset of over 100 million movie ratings from Netflix customers. 2) The authors initially approached the competition as a fun side project but ended up winning the $1 million prize as part of a team that achieved the greatest improvement over Cinematch. 3) The task was challenging because movies have many complex dimensions that influence ratings, and models need to capture the key factors without overfitting the data.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Netflix Prize: All Together Now: A Perspective On The

    Uploaded by

    fifi
    114 views1 page
    1) The document discusses the Netflix Prize competition, which challenged participants to improve Netflix's movie recommendation algorithm (Cinematch) by at least 10% based on a dataset of over 100 million movie ratings from Netflix customers. 2) The authors initially approached the competition as a fun side project but ended up winning the $1 million prize as part of a team that achieved the greatest improvement over Cinematch. 3) The task was challenging because movies have many complex dimensions that influence ratings, and models need to capture the key factors without overfitting the data.

    Original Description:

    ee

    Original Title

    s00144-010-0005-2

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    1) The document discusses the Netflix Prize competition, which challenged participants to improve Netflix's movie recommendation algorithm (Cinematch) by at least 10% based on a dataset of over 100 million movie ratings from Netflix customers. 2) The authors initially approached the competition as a fun side project but ended up winning the $1 million prize as part of a team that achieved the greatest improvement over Cinematch. 3) The task was challenging because movies have many complex dimensions that influence ratings, and models need to capture the key factors without overfitting the data.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    114 views1 page

    Netflix Prize: All Together Now: A Perspective On The

    Uploaded by

    fifi
    1) The document discusses the Netflix Prize competition, which challenged participants to improve Netflix's movie recommendation algorithm (Cinematch) by at least 10% based on a dataset of over 100 million movie ratings from Netflix customers. 2) The authors initially approached the competition as a fun side project but ended up winning the $1 million prize as part of a team that achieved the greatest improvement over Cinematch. 3) The task was challenging because movies have many complex dimensions that influence ratings, and models need to capture the key factors without overfitting the data.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 1

    All Together Now:

    A Perspective on the
    NETFLIX
    PRIZE
    Robert M. Bell, Yehuda Koren, and Chris Volinsky

    W
    hen the Netflix Prize was learning methods tend to center on their ratings on 18,000 movies. This
    announced in October of algorithms (black boxes), where the amounted to more than 100 million rat-
    2006, we initially approached focus is on the quality of predictions— ings. The task was to use these data to
    it as a fun diversion from our ‘day jobs’ at rather than ‘understanding’ what drives build a model to predict ratings for a
    AT&T. Our group had worked for many particular predictions. hold-out set of 3 million ratings. These
    years on building profiles of customer In contrast, statisticians tend to think models, known as collaborative filter-
    patterns for fraud detection, and we more in terms of models with parameters ing, use the collective information of
    were comfortable with large data sets, that carry inherent interest for explain- the whole group to make individualized
    so this seemed right up our alley. Plus, it ing the world. Leo Breiman’s article, “Sta- predictions.
    was about movies, and who doesn’t love tistical Modeling: The Two Cultures,” Movies are complex beasts. Besides
    movies? We thought it would be a fun which was published in Statistical Science, the most obvious characterization into
    project for a few weeks. provides various views on this contrast. genres, movies differ on countless dimen-
    Boy, were we wrong (not about the Our original team consisted of two stat- sions describing setting, plot, characters,
    fun part, though). Almost three years cast, and many more subtle features such
    isticians and a computer scientist, and
    later, we were part of a multinational as tone or style of the dialogue. The
    the diversity of expertise and perspec-
    team named as the winner of the $1 Movie Genome Project (www.jinni.com/
    million prize for having the greatest tive across these two disciplines was an
    important factor in our success. movie-genome.html) reports using “thousands
    improvement in root mean squared
    of possible genes.” Consequently, any
    error (RMSE) over Netflix’s internal
    finite model is likely to miss some of the
    algorithm, Cinematch. Fundamental Analysis
    The predominant discipline of par- signal, or explanation, associated with
    Challenge people’s ratings of movies.
    ticipants in the Netflix Prize appears
    to have been computer science, more The Netflix Prize challenge concerns On the other hand, complex models
    specifically machine learning. While recommender systems for movies. Net- are prone to overfitting, or matching
    something of a stereotype, machine flix released a training set consisting of small details rather than the big pic-
    data from almost 500,000 customers and ture—especially where data are scarce.
    24 VOL. 23, NO. 1, 2010

    You might also like