MindCoder: Automated and Controllable Reasoning Chain in Qualitative Analysis

Abstract: Extracting insights from qualitative analysis involves a series of reasoning steps, such as open coding, grouping, and identifying themes. We introduce the MindCoder reasoning chain, built on Chain-of-Thought (CoT) prompting, to support the insight extraction process step by step-including topic clustering, code labeling, conceptualization, and reporting. We designed the MindCoder web application to help users 1) automatically run this reasoning chain (i.e., obtain analysis report results in approximately 3-5 minutes) and 2) interactively control the reasoning process on demand. Our technical evaluations assess its reliability across various data types and demonstrate that simulated human iteration can potentially enhance coding quality. A user study further confirmed positive feedback regarding MindCoder's automation and its on-demand reasoning functionality.