Complete notes of the course

Page 1

Author: Paolo Roncaglioni Academic Year: 2019/20 Data Intelligence Applications Data Intelligence Applications Part 1: Pricing Economics Preliminaries Demand Curve Pricing Price discrimination Single product demand curves Interdependent product demand curves Strategic dependency Learning the demand curve A/B/n Testing A/B testing (sequential) A/B/n testing Weaknesses Testing technicalities Bandit algorithms Regret Minimization Definition Comparisons Technicalities UCB1 Thompson Sampling Delayed feedbacks How many candidate per-arm we should use? Non-stationary environments Sliding window Contextual generation Formal model Feature tree Part 2: Matching Alternating-path algorithms Hopkroft-Karp Weighted alternating path Combinatorial bandits and matching problems Thompson Sampling pseudocode Hungarian algorithm Online matching problems Greedy algorithm Randomized algorithm Generalization (weighted) Postponed Dynamic Deferred Acceptance Part 3: Advertising Introduction Pay per click advertising Payments Auctions Display advertising Sequence of events Second price Auction Optimization problem (advertiser) Model Algorithm Regression by Gaussian Processes Kernel functions Combinatorial GP bandits Non-combinatorial Combinatorial For advertising Part 4: Social influence Information effects Cascade effect Direct effects Seeding Extensions Influence maximization General case of the algorithm Monte Carlo sampling Exact solution Greedy algorithm Learning issues in influence maximization Scenarios Linear UCB Part 1: Pricing Economics Preliminaries Basic scenario: Single seller (monopoly) Infinite units of single goods Multiple customers Seller model: The seller has a cost over the good The seller can set a unique price for all the units of the good Customer model: Every customer…