This paper introduces VIDEE, a system that empowers entry-level data analysts to perform advanced text analytics using intelligent agents. VIDEE implements a human-agent collaborative workflow and consists of (1) a decomposition phase that incorporates a human-in-the-loop Monte-Carlo Tree Search algorithm to support generative inference using human feedback; (2) an execution phase that generates executable text analytics pipelines; and (3) an evaluation phase that integrates LLM-based evaluation and visualization to support user validation of results. In this study, we conducted two quantitative experiments to evaluate the effectiveness of VIDEE and analyze agent errors. Furthermore, we conducted user studies with participants of varying skill levels, from those with little to no NLP and text analytics experience to experts, to demonstrate the usability of the system and identify user behavior patterns.
