Year: 2024
Type: dissertation
Abstract: Digital media is one of the cornerstones of modern product marketing. Online display campaigns are one example of such media. With accurate machine learning models, we can leverage user contextual information and online activity to further target these display campaigns at a user level, making them more compelling. This will further improve the main business KPIs, such as click-through rate and customer conversion, thereby driving revenue growth. This master's thesis investigates the application of reinforcement learning, more specifically a contextual bandit algorithm, for optimizing online advertisement placement on the web pages of Sanoma Media Finland. In this project, we create a minimum viable solution to explore the feasibility and accuracy of this approach in practice. We implement this using the efficient open-source library \textit{Vowpal Wabbit} and create a supporting architecture for it in AWS, allowing the solution to be further scaled up if taken into production. Due to practical constraints, the model's performance was assessed using historical data rather than live online data. The results showed a significant potential of the model to improve key metrics like click-through rate for various products, with notable improvements in certain cases. However, the study acknowledges the limitations of using historical data and suggests the need for further testing in a real-time environment to validate these findings and to assess the model's impact on subscription rates.
Source: Aaltodoc (Aalto University)
Author: Veneranta, Leevi
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7060866916