1. Introduction
Correlation/Prediction ≠ Causation
● Causation disentangles the relationship among multiple factors.
● Causation helps understand the underlying mechanism.
● Causations allows business decisions to involve counterfactual reasoning.
Sequential Decision-making Problem
Multi-armed Bandit Problem
The right product for the right customer at the right moment (i.e. e-commerce).
● Suppose you want to play the “best” machine. Use the one with the highest payoff →
Exploit current knowledge.
● Suppose you want to learn more. Machine B might be more profitable than Machine
A if you play it more → Explore more.
Bandit Framework
Multi-armed Bandit Contextual Bandit
Reinforcement Learning
Offline Policy Evaluation
Uplift Modeling
Improve return on investment (ROI) by
targeting only the persuadable customers.
The goal is to rank users/customers.
1
, Marketing Basic Tenant
● Action should generate greater value for the targeted audience.
● Inform customer-oriented decision-making on product, pricing, promotion, and place.
● Appropriate deployment of marketing resources.
▶ Marketing as a research field is an integration of multiple research fields (i.e. econometrics,
psychology, etc.).
▶ Data Science is increasingly used in the field of marketing.
Big Data (In Marketing)
● Defined by the 3 V’s:
○ Volume (amount).
○ Velocity (speed i.e. batch or stream).
○ Variety (type i.e. structured or unstructured).
○ Veracity (accuracy).
○ Variability (inconsistency).
○ Visualization (ways to visualize).
○ Value (business value).
○ Volatility (storage & retrieval).
● Problem recognition:
○ Detect customer dissatisfaction.
○ Provide info for development & improvement.
○ Dialogue, shopping, and use behaviors.
○ Change in transaction pattern.
○ Comments, complaints, opinions on social media.
● Search and alternative evaluation:
○ Understand how certain search terms relate to their products or services to
improve keyword-specific-click through rate (CTR).
● Post consumption evaluation/engagement:
○ Product reviews/user feedback can be used to infer satisfaction, loyalty, or
commitment.
▶ Big data is the result of digitalization (more informative than its physical size or complexity).
Query Variation Index
Whether a keyword allows identification of a user's information needs → good predictor
keyword performance.
2
Correlation/Prediction ≠ Causation
● Causation disentangles the relationship among multiple factors.
● Causation helps understand the underlying mechanism.
● Causations allows business decisions to involve counterfactual reasoning.
Sequential Decision-making Problem
Multi-armed Bandit Problem
The right product for the right customer at the right moment (i.e. e-commerce).
● Suppose you want to play the “best” machine. Use the one with the highest payoff →
Exploit current knowledge.
● Suppose you want to learn more. Machine B might be more profitable than Machine
A if you play it more → Explore more.
Bandit Framework
Multi-armed Bandit Contextual Bandit
Reinforcement Learning
Offline Policy Evaluation
Uplift Modeling
Improve return on investment (ROI) by
targeting only the persuadable customers.
The goal is to rank users/customers.
1
, Marketing Basic Tenant
● Action should generate greater value for the targeted audience.
● Inform customer-oriented decision-making on product, pricing, promotion, and place.
● Appropriate deployment of marketing resources.
▶ Marketing as a research field is an integration of multiple research fields (i.e. econometrics,
psychology, etc.).
▶ Data Science is increasingly used in the field of marketing.
Big Data (In Marketing)
● Defined by the 3 V’s:
○ Volume (amount).
○ Velocity (speed i.e. batch or stream).
○ Variety (type i.e. structured or unstructured).
○ Veracity (accuracy).
○ Variability (inconsistency).
○ Visualization (ways to visualize).
○ Value (business value).
○ Volatility (storage & retrieval).
● Problem recognition:
○ Detect customer dissatisfaction.
○ Provide info for development & improvement.
○ Dialogue, shopping, and use behaviors.
○ Change in transaction pattern.
○ Comments, complaints, opinions on social media.
● Search and alternative evaluation:
○ Understand how certain search terms relate to their products or services to
improve keyword-specific-click through rate (CTR).
● Post consumption evaluation/engagement:
○ Product reviews/user feedback can be used to infer satisfaction, loyalty, or
commitment.
▶ Big data is the result of digitalization (more informative than its physical size or complexity).
Query Variation Index
Whether a keyword allows identification of a user's information needs → good predictor
keyword performance.
2
