Aaron P. Kaye : The Personalization Paradox

👉 Speaker : Aaron P. Kaye : The Personalization Paradox : Welfare Effects of Personalized Recommendations in Two-Sided Digital Markets

00:00 Introduction
00:20 Recommendation systems determine product rankings
01:05 What is a Two-Sided Digital Market
03:15 Introduction
04:35 Exemple : The Elvis Hotel
09:15 Price Competition for Product Rankings
10:10 Co-Ranking of Close Substitutes
10:40 This paper
12:50 Preview of Results
14:40 Contributions
15:05 Outline
15:30 Data and Setting
15:35 Data : Expedia Competition to Personalize Recommendtions
16:35 Ecommerce Platform Design : Recommendations and Feature Empire
17:20 Position Effects and Incentives
17:25 Empirical Evidence of Position Effects
17:35 Slot impacts demand even when recommendations are random
18:25 Empirical Evidence of Position Effects
18:40 Slot is correlated with price and hidden product features
19:10 Empirical Evidence of Position Effects
20:00 Structural Model
20:05 Structural Model Outline
23:00 Structural Model : Demand
23:05 Demand Model
24:10 Demand Model : Model Details
25:15 Demand Estimation Details (Maximum Simulated Likelihood)
26:05 Demand Estimation : Utility, Search Cost, and reservation Utility
27:05 Consumer Choice Model Identification
28:30 Demand Results
28:45 Structural Model : Plateform Recommendations (expedia)
29:00 Model : Platform Two Step "Model of the Model" Approach
30:20 Platform Results - Out of Sample Fit
30:50 Platform Model Sequential Logit Results
30:55 Structural Model : Supply (Hotels)
31:00 Model : Supply Side
32:10 Supply Side Estimation : Two Stage Least Squares
33:05 Supply Side Results
33:30 Personalized Recommendation Systems Training for Counterfactuals
33:40 Recommendations Systems
34:10 Recommendations Systems (Ensemble of LambdaMARTs)
36:10 Counterfactuals
36:15 Counterfactual Setup
36:25 Results
37:35 Personalized recs. With Star-level economies of scale and softy capacity
38:20 Counterfactual Results Continued
39:45 Policy Counterfactual : Price Tuned Recommendations
40:25 Paper Overview
41:20 Conclusion
42:00 Discussion