Learning from big data

Every day, millions of consumers voice their opinions in product-review websites, blogs and chat rooms. They produce massive amounts of user-generated content (UGC), most of which is textual and freely available for analysis. At the same time, online retailers personalize their websites and tailor how they make recommentations to, communicate with, and advertise to individual consumers. This generates massive and rich datasets that contain valuable information about individual customers. All that can and must be done within legal frameworks, such as GDPR.
In this era of big data, the availability of larger and more diversified data sets opens up exciting opportunities for ethical marketing researchers and practitioners. However, with so much data available, it can be hard to determine what information is correct or how to separate data from information. For example, how can you extract information about consumer preferences from reviews and blogs? And how do you go from data to information to insights and to marketing decisions?

In our current era, we find ourselves inundated with an abundance of data, yet paradoxically, we often have to make decisions with limited information. Picture this: a company introducing a new product to the market. They face a dilemma – do they stick with tried-and-tested marketing strategies, or do they venture into uncharted territory to explore potentially lucrative opportunities? Similarly, in online advertising, businesses must decide whether to invest in familiar ad placements or experiment with new platforms to reach untapped audiences. These scenarios underscore the critical need for a nuanced approach to decision-making, one that balances the certainty of past successes with the uncertainty of future learnings. Multi-armed bandits (MABs) provides a framework for optimizing decisions in situations where there is a trade-off between exploiting known strategies (the certainty of past successes) with exploring new options (uncertainty of future learnings). By dynamically allocating resources to explore various alternatives while simultaneously exploiting the most promising options, MABs offer a data-driven solution to the challenge of decision-making in uncertain environments. In the context of the examples mentioned earlier, MAB algorithms can help businesses efficiently allocate advertising budgets, identify optimal pricing strategies, and maximize returns on new product investments.

All that is needed to address these challenges is the right set of tools and the training that helps you to use these tools correctly. This course first teaches how to formalize marketing problems as statistical models. It then shows how to solve marketing problems by complementing econometric techniques with two types of modern machine learning (ML) methods: natural language processing (NLP) and multi-armed bandits (MABs). In short, you will find tools and conceptual frameworks needed to identify and exploit the opportunities that big data sources open up. The course has two modules, each containing lectures and hands-on activities:

• Mine your Own Business in Blogs, Reviews and Tweets
• Mastering sequential decision-making with multi-armed bandit models (MABs)

• Formulate marketing decisions as machine learning tasks.
• Explore the fascinating world of multi-armed bandits (MABs), essential tools that are reshaping decision-making in various industries, and learn how MABs drive personalized recommendation systems, dynamic pricing strategies, and effective ad allocation for startups and major BigTech companies.
• Extract relevant insights on consumer preferences and behavior from user-generated content (UGC).
• Collaborate efficiently when working on data science projects in teams.
• Infer machine learning best practices from academic literature.

This course relies a lot on statistical concepts and tools you are expected to apply independently (or study independently to catch up with gaps in your knowledge). Successful participation in at least one advanced statistics or data science courses is strongly recommended (e.g., Applied Statistics, Econometrics 1, Introduction to Multivariate Statistics). Also note that other data science, statistics, and computer science courses might cover more specific knowledge that can be useful for this course.

During the lectures, the teachers will use R to demonstrate how to implement the machine learning models taught in the course. In the homework assignments, you will write your own machine learning pipelines in R (data transformation, model estimation, model validation), so you must be proficient in R before starting this course or a in similar programming language (such as Python). All course material is written in R. Please keep in mind that this course will not teach you how to use R, so please prepare in advance. Consider using time in the Summer to sharpen your R skills before the start of the course.

Teaching methods

• Lecture
• Tutorials
• Group assignment and individual assignments
• All RSM minors have mandatory attendance

Teaching materials
Details will be announced at the start of the course.

Method of examination

• One group assignment, in groups of 4 to 5 students
• Two individual assignments
• Participation in lectures (attendance, responses to cold calls, other contributions in class)

Composition final grade

• Group assignment (30%)
• Individual assignments (60%)
• Participation (10%)