Further information

Background
Modern websites often have a dynamic structure, i.e. their content is constantly changing due to server-side updates, user interactions or personalized deliveries. This poses a particular challenge when certain content needs to be documented and tracked - especially in the context of possible legal violations (e.g. prohibited advertising, illegal content).
In order to document violations, a screenshot is typically taken showing the content in question, together with a description of the violation. If several screenshots are taken of the infringement or the website, these must be compared manually.

Objective of the work
The aim of the work is the development and evaluation of AI-supported methods for the automated comparison of dynamic websites (with a focus on screenshots). It will be investigated whether multimodal LLMs are suitable for automating this process and supporting the detection and comparison of such violations.