Designing AI-Based Systems for Qualitative Data Collection and Analysis

With the continuously increasing impact of information systems (IS) on private and professional life, it has become crucial to integrate users in the IS development process. One of the critical reasons for failed IS projects is the inability to accurately meet user requirements, resulting from an incomplete or inaccurate collection of requirements during the requirements elicitation (RE) phase. While interviews are the most effective RE technique, they face several challenges that make them a questionable fit for the numerous, heterogeneous, and geographically distributed users of contemporary IS.

Three significant challenges limit the involvement of a large number of users in IS development processes today. Firstly, there is a lack of tool support to conduct interviews with a wide audience. While initial studies show promising results in utilizing text-based conversational agents (chatbots) as interviewer substitutes, we lack design knowledge for designing AI-based chatbots that leverage established interviewing techniques in the context of RE. By successfully applying chatbot-based interviewing, vast amounts of qualitative data can be collected. ... mehrSecondly, there is a need to provide tool support enabling the analysis of large amounts of qualitative interview data. Once again, while modern technologies, such as machine learning (ML), promise remedy, concrete implementations of automated analysis for unstructured qualitative data lag behind the promise. There is a need to design interactive ML (IML) systems for supporting the coding process of qualitative data, which centers around simple interaction formats to teach the ML system, and transparent and understandable suggestions to support data analysis. Thirdly, while organizations rely on online feedback to inform requirements without explicitly conducting RE interviews (e.g., from app stores), we know little about the demographics of who is giving feedback and what motivates them to do so. Using online feedback as requirement source risks including solely the concerns and desires of vocal user groups.

With this thesis, I tackle these three challenges in two parts. In part I, I address the first and the second challenge by presenting and evaluating two innovative AI-based systems, a chatbot for requirements elicitation and an IML system to semi-automate qualitative coding. In part II, I address the third challenge by presenting results from a large-scale study on IS feedback engagement. With both parts, I contribute with prescriptive knowledge for designing AI-based qualitative data collection and analysis systems and help to establish a deeper understanding of the coverage of existing data collected from online sources. Besides providing concrete artifacts, architectures, and evaluations, I demonstrate the application of a chatbot interviewer to understand user values in smartphones and provide guidance for extending feedback coverage from underrepresented IS user groups.