User research aims to understand users' needs, behaviors, and attitudes to effectively inform the design and development of products or services. It is a key endeavor to learn how users experience digital technologies, what is working well and what is not, and identify gaps and future needs in order to personalize and improve the user experience. To design for positive user experiences, investigating affective user reactions (e.g., emotions, stress, flow) is of particular interest. Therefore, affective user research collects and analyzes behavioral data and affective reactions of users when engaging with products or services. With the growing amount of data and computing capabilities, artificial intelligence (AI) technologies are increasingly used in user research for the prediction of affective user states when interacting with digital technology.

The recent advances in artificial intelligence (AI), however, may not only support affective user research as a method of inquire, but it also has found its way into our daily lives as humans interact with it every day, for example, in form of recommendation engines on social media, in health applications, or as personal assistants based on large language models (LLMs) to receive text output for code completion, ideation, or writing. Interacting with AI-based digital technologies also triggers affective user reactions. However, these affective user reactions in Human-AI Interactions are yet to be understood. In this seminar, participants will apply methods for affective user research on a particular type of Human-AI Interaction, the prompting of LLMs. LLM prompting is expected to become the up-and-coming form of interacting with AI in the future. To receive output from an LLM, users must send a prompt to the LLM. Given a prompt, an LLM responds incrementally with “tokens” (e.g., groups of letters, numbers, punctuation) which build the output. Structuring the prompt and receiving output influences the affective reactions of the user. Precisely, these user reactions should be investigated by the students participating in this seminar.

In the "Affective User Research for Human-AI Interaction" seminar, participating students will learn how to apply AI-based user research methods with a specific emphasis on the affective dimension when interacting with AI-based digital technologies.

The goal of this seminar is to provide students with a unique set of skills in (1) quantitative data analysis, (2) knowledge about Human-AI Interaction and, in particular, LLM prompting, and (3) prediction of affective user states (e.g., emotions, stress) using state-of-the-art machine learning (ML) techniques. Students will leverage a dataset on Human-AI Interaction and gain in-depth knowledge from it as part of the seminar. The seminar emphasizes the importance of applying the aforementioned affective user research methods in an ethically compliant form.

The core activities include:

• Learn the fundamentals of AI-based affective user research methods.
• Explore a dataset on Human-AI Interaction with the specific focus on the interplay of user behavior and affective user reactions.
• Developing AI-based supervised machine learning techniques for predicting user activities and affective user states.
• Present findings and insights to the seminar audience and discuss the results

The seminar is held by Dr. Ivo Benke (Biontech) in cooperation with Dr. Lennard Schmidt (Google). Both are experts from industry in the fields of affective user research, quantitative data analysis, and Human-AI Interaction.

Learning Objectives

• Understand the potential of combining user behavior and affective user reaction data for affective user research.
• Develop hands-on knowledge by applying AI-based affective user research methods on a real-world dataset.
• Develop a deeper understanding of a prominent form of Human-AI Interaction (e.g., LLM prompting).
• Deliver a presentation in a scientific context in front of an auditorium.

Language of instruction: English