b9612eab06af8cf383e27681bf076b56.ppt
- Количество слайдов: 167
What is “Computational Advertising”?
Transparency and value
Extract Topical info Increases coverage, more relevant match
Precision Recall 25% lift in precision at 10% recall
Retrospective data [URL, ad, is. Clicked] Crawl URLs a sample of URLs Classify pages and ads Rare event estimation using hierarchy Impute impressions, fix sampling bias
Retrospective data [page, ad, isclicked] Crawl Pages a sample of pages Classify pages and ads Rare event estimation using hierarchy Impute impressions, fix sampling bias
Unobserved “state”
om d TS an R S LM , N
Bandit “arms” (unknown payoff probabilities)
Bandit “arms” (unknown payoff probabilities)
Priority 1 Priority 2 Priority 3
Bandit “arms” = ads
One bandit Unknown CTR Content Match = A matrix
Root Apparel Computers Travel Page/Ad
…… … … …… Consider only two levels
Ad parent classes Ad child classes …… … … …… Consider only two levels Block One bandit
Ad parent classes Ad child classes …… … … …… Block One bandit
?
ad ?
ad
# clicks in cell # impressions in cell All cells in a block come from the same distribution
Estimated CTR Beta prior (“block CTR”) Observed CTR
Root 20 nodes … 221 nodes ~7000 leaves Taxonomy structure We use these 2 levels
Clicks Number of pulls Multi-level gives much higher #clicks
Number of pulls Mean-Squared Error
Mean-Squared Error Clicks Number of pulls


