Introduction to contexual bandit

  • Published on
    15-Jun-2015

  • View
    2.877

  • Download
    1

Embed Size (px)

Transcript

  • 1. Contextual Bandit @DSIRNLP m.tsubosaka@gmail.com

2. BanditContexutal Contextual Bandit 3. Bandit CTR Finite-time Analysis of multiarmed bandit problem, Machine Learning,2002 4. CTR 100%CTR8.2% CTR 8.2%CTR 5% 5. 80%20% CTR9%CTR 10% CTR 1%CTR 5% CTR 5% 6. Bandit arm Bandit 7. Contextual bandit armcontext context CTR LinUCB CTR = 0.1 * + 0.01 * CTR = 0.05 * + 0.05 * 8. LinUCB() context+Upper confidence 9. 70%, 30% 1: CTR 10%, CTR 2% 2: CTR 2%, CTR 10% context (1,0), (0,1) 10. 1100 contextLinUCB CTR CTRUCB7.56%LinUCB10.0% 11. A contextual-bandit approach to personalized news article recommendation, WWW 2010 LinUCB An empirical evaluation of thompson sampling, NIPS 2011 Content recommendation on web portal, CACM 2013