Рекомендательная система: быстрый старт

Embed Size (px)

DESCRIPTION

Lecture for DM labs, first part. Introducing iALS and demonstrating ORYX recommender

Text of Рекомендательная система: быстрый старт

  • 1. :

2. ! 3. RS Available data User history Content Collaborative Content-based Hybrid Tags & Metadata 4. , 5. 6. ? (explicit feedback) ? Implicit feedback ! feedback- , 7. iALS min x*,y* cui (pui xu T yi ) u,i 2 + xu u 2 + yi i 2# $ % & ' ( 8. iALS (1) min x*,y* cui (pui xu T yi ) u,i 2 +... 9. iALS Preferences User features Item features x # users #items # users #items # features #features 10. iALS (2) min x*,y* cui (pui xu T yi ) u,i 2 + xu u 2 + yi i 2# $ % & ' ( 11. ? 12. Alternating Least Squares xu = YT Cu Y + I( ) 1 YT Cu p(u) yi = YT Ci Y + I( ) 1 YT Ci p(i) X-step: Y-step: 13. Alternating Least Squares xu = YT Cu Y + I( ) 1 YT Cu p(u) YT Cu Y =YT Y +YT Cu I( )Y X-step: 14. Alternating Least Squares (2) xu = YT Cu Y + I( ) 1 YT Cu p(u) X-step: Au du Au = A0 + c0 +cui( )Yu T Yu i:(u,i)R du = d0 + c0r0 +cuirui( )Yu i:(u,i)R 15. Oryx 16. Oryx 17. Audioscrobbler dataset (2005) 148 111 1 631 028 24 296 858 http://www-etud.iro.umontreal.ca/~bergstrj/audioscrobbler_data.html 18. ! 19. apdanilchenko@yandex.ru !