05.11.2025Semester Award for Kevin Bründler

Email remains a highly relevant communication medium – around 300 billion emails are sent every day. Many of them contain valuable business information, such as contract details, prices, or delivery details. However, this information is often not available in standardised formats and therefore not easily accessible for companies to process.

In his Master's Thesis “Exploring Text Mining Applications in Financial Negotiations; A Case Study on Synthetic Data”, Kevin Bründler explores how information can be extracted from emails in a structured form without relying on real, confidential business data, using the maritime chartering industry as a testbed. Kevin developed a pipeline to generate synthetic emails with which NLP (natural language processing) models can be trained, thus protecting privacy and business secrets – companies are hesitant to use real, confidential data for the training of NLP models.

The results from Kevin’s work have applications in the maritime chartering industry but can also be applied to other industries such as finance, logistics, or business-to-business distribution. These are domains where it is difficult for external companies to provide solutions based on NLP models without having access to appropriate training data. Synthetic training data provides a solution to this dilemma.

Kevin’s thesis also presents empirical findings on the quality of different methods for generating training data. He shows the limitations of current Large-Language Models, and evaluates different extraction architectures in a real-world setting.

While studying Data Science and Informatics, Kevin Bründler worked in a trading company that transports goods worldwide as a charterer. There he experienced firsthand how valuable the information in emails is – and how difficult it is to evaluate it in a systematic way.