Presentation is loading. Please wait.

Presentation is loading. Please wait.

Collaborative Filtering with Temporal Dynamics Yehuda Koren Yahoo! Israel KDD 2009.

Similar presentations


Presentation on theme: "Collaborative Filtering with Temporal Dynamics Yehuda Koren Yahoo! Israel KDD 2009."— Presentation transcript:

1 Collaborative Filtering with Temporal Dynamics Yehuda Koren Yahoo! Israel KDD 2009

2 Where it all begins … 很久很久以前

3 Where it all begins … 很久很久以前 (2006)

4 Where it all begins … 很久很久以前 (2006)

5 Where it all begins …

6

7 Given (Kevin, Avatar, 2009/12/20, ★ ★ ★ ★ ★ ) (Coca, 2012, 2009/12/10, ★ ★ ★ ★ ) Predict (Kevin, District 9, 2009/12/18, ?????)

8 Corpus Training Dec 31, 1999 – Dec 31, 2005 100 million ratings 500 thousand users 17,770 movies Testing 1.4 million ratings

9 How do I participate?

10

11 The $1,000,000 Winner is … BellKor ’ s Pragmatic Chaos

12 The $1,000,000 Winner is … BellKor ’ s Pragmatic Chaos

13 The $1,000,000 Winner is … BellKor ’ s Pragmatic Chaos

14 The $1,000,000 Winner is … BellKor ’ s Pragmatic Chaos Koren Bell = Bob Bell Kor = Koren

15 圖解 The BellKor Method Temporal

16 圖解 The BellKor Method Temporal UserMovie

17 圖解 The BellKor Method Temporal UserMovieInteraction

18 The Bellkor Function

19 User ’ s rating of a movie

20 The Bellkor Function Average movie rating of all movies

21 The Bellkor Function Movie Temporal Effect

22 The Bellkor Function User Temporal Effect

23 The Bellkor Function Interactive Temporal Effect between Users and Movies

24 Movie Temporal Effect

25 Time-independent parameter

26 Movie Temporal Effect Time-dependent parameter; grouped into intervals of 10 weeks

27 User Temporal Effect

28 User-independent parameter

29 User Temporal Effect Temporal trend

30 User Temporal Effect Daily variability

31 User-Item Temporal Effect

32 Matrix Factorization timeSVD++

33 User-Item Temporal Effect Movie vector

34 User-Item Temporal Effect Temporal user vector

35 User-Item Temporal Effect Movies that the user has rated

36 Computing the Parameters Minimize the squared error of the function over training data

37 Performance on Quiz Set Netflix ’ s Cinematch: 0.9514 BellKor: 0.8799

38 This is only the beginning …


Download ppt "Collaborative Filtering with Temporal Dynamics Yehuda Koren Yahoo! Israel KDD 2009."

Similar presentations


Ads by Google