Problem Description

In an earlier conducted experiment, we collected dialogs between participants and ChatGPT. The participants were randomly attributed to one of two treatments: human-like and computer-like. Afterwards we argued about the effects of this chage and if particpants might behave in a different way as anticipted: e.g. talking about different topics in one of the treatments. In addition we detected a subgroup in both treatments and we are now interested if we could detect them by analyszing only the dialog. The data will be provided.

Goal of the Thesis

The goal of the thesis is to analyze the text data and test if patterns can be found that were not anticipated.

In your thesis, you should start with a brief literature review of experiments and unintentionally introduced biases. This should be followed by a description of the problem and a methods chapter that describes different ways to detect the downstream effects of potential introduced biases and the aforementioned subgroup. The main part will be your use of the method and your results, showing whether you could detect a potential bias or subgroup.

Requirements

• Proficiency in Python
• Natural Language Processing
• The ability to use Git