Infomediary competition

Abstract/Contents

Abstract

Acknowledgments I am deeply grateful to my dissertation adviser, Prof. Samuel Chiu, who mentored, coached, and taught me, and without whom this dissertation could not be possible. I am greatly indebted to Prof. Yinyu Ye, who has championed me, and given me opportunities to grow as both a teacher and a researcher. And many thanks also go to my dissertation committee, Amin Saberi as well as Edison Tse, who served as great instructors to me, whose research inspired me, and whose assistance greatly improved this dissertation effort. I am grateful to Dr. Gio Wiederhold also for graciously chairing my defense. And I would like to thank Lori Cottle for her support. And lastly, I would like to thank my parents, Brian Tolman and Lisa Shepard, whose good examples motivated me to undertake and finish this rich endeavor. Abstract This dissertation studies the competitive dynamics in the infomediary market—where infomediaries are defined as: those entities that view themselves as businesses who connect two-sided markets, and whose primary benefit offered is collecting and organizing the information accumulated from the interactions of the two sides through their provided platform, for the benefit of the clients on both sides and the efficacy of their interactions. We study the information-intermediary market under a comprehensive conceptual framework, and show that the these simple platform businesses—which are often lacking infrastructure costs—have fewer competitive factors at play than those businesses in more traditional markets. We show that the lack of traditional stabilizing competitive forces suggest winner-take-all dynamics will be found in the infomediary market. We seek to further analyze the competitive factors of the infomediary market by building a repre- sentative model—an optimal-control dynamic system model solved through dynamic programming. We use this model to study specific challenger vs. incumbent scenarios—often pulling example scenarios from well-known infomediaries that entered the market against a challenger. The pre- dicted outcomes of these scenarios are analyzed and can be compared to the actual outcomes of the real-world situation they were based upon. From these predictions we also draw conclusions and hypothesize unique strategies for success—for both entering challengers and responding incumbents. One main competitive factors identified is the switching cost incurred by users when they leave an incumbent infomediary. The data the incumbent collected pertaining to these users empower the incumbent to perform more predictive collaborative filtering, which empowers them to make more useful recommendations to the users. In order to understand this factor better, we survey all major recommender-system techniques, discuss and analyze them. In order to draw conclusions about infomediaries as a whole, we seek to unify the major collabo- rative filtering methods under one newly-defined umbrella class of models. we successfully show that most methods can be viewed under one theoretical and computational umbrella. We iteratively show how each method can be brought into this framework, and then use axioms of Bayesian Probability to discuss under what precise circumstances each method would or would not be most appropriate. We connect the analysis of collaborative filtering with the analysis of switching costs through a value-of-information analysis performed upon a recommender system model that we justify as being suitably generic. This analysis allows us to assert how an infomediary's incumbent advantage will grow over time. We compare these assertions to existing research in both the business, machine learning, and operations research fields, draw strategic conclusions for both incumbents and new ventures in the infomediary market. Contents 0 Introduction 0.1 Background 0.2 Main Contribution 1 Collaborative Filtering Techniques 1.1 Nearest-Neighbor Methods 1.2 Least-Squares Factoring 1.3 Factoring with Regularizers 1.4 Probabilistic Latent Semantic Analysis 1.5 Wishart Normal Factor Model 1.6 Other Models 2 Theoretical Insights from a Related Problem 2.1 Sparse Selection 2.2 Estimate Confidence 2.2.1 Sparse Observations 3 Probability Matrix Factorization Models 3.1 The Bayesian Wishart-Normal 3.2 PMF Class 13 3.3 Relation to Linear Regression 3.4 Relation to Ordinary Least Squares 3.5 Relation to Latent Semantic Analysis 3.6 Relation to Nearest Neighbor Models 3.6.1 Nearest Neighbor Captures a Converging Characteristic of the PMF Models . 3.6.2 PMF Models Can Capture Preferences Characterized by Nearest Neighbor . 3.6.3 In Practice the Two Models are Incompatible 3.6.4 Improvements to the Nearest Neighbor Models Already Exist in PMF Models 3.6.5 Nearest Neighbor Model are not Robust to the Independently Sampling As- sumption 3.7 Relation to Cluster Models 3.8 Solving the Full PMF Posterior 3.8.1 An Example with the Wishart Normal 3.9 Justifying the choice of PMF 3.10 Easily Blended with Content Models 3.11 Summary 34 4 Sensitivity Analysis—the Value of Information 4.1 Value of Information 4.2 The Recommender Model 4.3 Results 4.4 Trends 5 Early-Mover Versus Late-Mover Advantages 5.1 First-Mover Background 5.2 Economic Factors 5.2.1 Entry Scale 5.2.2 Demand Uncertainty 5.2.3 Scale-to-Market Size 5.2.4 Response Time 5.2.5 Advertising Intensity 5.2.6 Scope Economies 5.3 Preemption and Technological Factors 5.4 Behavioral Factors 5.4.1 Nature of Good 5.4.2 Market Type 5.4.3 Market Evolution 5.4.4 Buyers' Investment in Cospecialized Assets 5.5 Latecomer Advantages 5.5.1 Free-Rider Effects 5.5.2 Learning from Strategic Mistakes 5.6 Summary of Competitive Advantages 6 Competitive Strategy 6.1 Users' Infomediary Choice 6.1.1 Collective vs. Individual Evaluation 6.1.2 Multi-Armed Bandit 6.1.3 Modeling the Infomediary Uncertainties 6.2 The Dynamic Competition Model 6.2.1 Users' Decision Model 6.3 Challenger with Equal Yet Uncertain Learning Rate 6.3.1 Stronger Incumbent 6.3.2 Inflated Challenger 6.4 Challenger with Higher Yet Certain Learning Rate 6.5 Modeling Distinct User Types 6.6 Expanding User Base 6.7 Highly Evolving Markets 6.8 Expanding Platforms 6.9 Simple Preferences 6.10 Advertising Intensity 6.10.1 Monetization Strategy 6.10.2 Incentives to Merge 6.11 Dynamic Model Conclusions 7 Conclusions 7.1 Increasing Accuracy 7.2 Using Separation as a Tool for Trust 7.3 Product Development is a Potential Opportunity for a Two-Sided Market 7.4 Comprehensive Challenger Strategy 7.5 Experimenting with Monetization Approaches 8 Further Research 8.1 Monetization of Media 8.2 Expanding the Definition of an Infomediary 8.3 Empirical Justification END ABSTRACT.

Description

Type of resource	text
Form	electronic; electronic resource; remote
Extent	1 online resource.
Publication date	2014
Issuance	monographic
Language	English

Creators/Contributors

Associated with	Tolman, Caleb
Associated with	Stanford University, Department of Management Science and Engineering.
Primary advisor	Chiu, Samuel S
Thesis advisor	Chiu, Samuel S
Thesis advisor	Saberi, Amin
Thesis advisor	Ye, Yinyu
Advisor	Saberi, Amin
Advisor	Ye, Yinyu

Subjects

Genre	Theses

Bibliographic information

Statement of responsibility	Caleb Tolman.
Note	Submitted to the Department of Management Science and Engineering.
Thesis	Thesis (Ph.D.)--Stanford University, 2014.
Location	electronic resource

Access conditions

Copyright

© 2014 by Caleb Joseph Tolman

Also listed in

View in SearchWorks

Loading usage metrics...