Jupybara: Operationalizing a Design Space for Actionable Data Analysis and Storytelling with LLMs

Abstract: Mining and conveying actionable insights from complex data is a key challenge of exploratory data analysis (EDA) and storytelling. To address this challenge, we present a design space for actionable EDA and storytelling. Synthesizing theory and expert interviews, we highlight how semantic precision, rhetorical persuasion, and pragmatic relevance underpin effective EDA and storytelling. We also show how this design space subsumes common challenges in actionable EDA and storytelling, such as identifying appropriate analytical strategies and leveraging relevant domain knowledge. Building on the potential of LLMs to generate coherent narratives with commonsense reasoning, we contribute Jupybara, an AI-enabled assistant for actionable EDA and storytelling implemented as a Jupyter Notebook extension. Jupybara employs two strategies -- design-space-aware prompting and multi-agent architectures -- to operationalize our design space. An expert evaluation confirms Jupybara's usability, steerability, explainability, and reparability, as well as the effectiveness of our strategies in operationalizing the design space framework with LLMs.

• The paper introduces Jupybara, a Jupyter Notebook extension using LLMs to operationalize a design space for actionable exploratory data analysis and storytelling.
• Jupybara's design space guides LLMs through semantic, rhetorical, and pragmatic dimensions to ensure accurate, persuasive, and relevant data insights.
• Key features like design-space-aware prompting and multi-agent architectures, integrated into Jupyter, resulted in user preference over standalone LLMs in studies.

The paper "Jupybara: Operationalizing a Design Space for Actionable Data Analysis and Storytelling with LLMs" introduces Jupybara, an AI-enabled assistant designed to facilitate actionable exploratory data analysis (EDA) and data storytelling. Implemented as a Jupyter Notebook extension, Jupybara leverages the capabilities of LLMs for generating coherent narratives and commonsense reasoning in data analysis and storytelling.

The core contribution is a design space framework that includes semantic, rhetorical, and pragmatic dimensions:

1. Semantic Precision focuses on accurately interpreting, monitoring, and verbalizing data findings.
2. Rhetorical Persuasion pertains to strategically mining and communicating analytical results to consolidate actionable insights.
3. Pragmatic Relevance integrates domain expertise with findings to derive actionable insights.

Key features of Jupybara include:

• Design-Space-Aware Prompting: A strategy to guide LLMs using a comprehensive set of guidelines based on the design space. This ensures that LLMs deliver interpretations, strategies, and insights that align with the principles of the design space.
• Multi-Agent Architectures: This involves employing multiple LLM agents for EDA and storytelling tasks. Agents collaborate to refine tasks iteratively, enhancing the alignment with each dimension of the design space and improving the quality of responses.
• Notebook Integration: Built as a Jupyter Notebook extension, Jupybara seamlessly integrates into existing data analysis workflows, allowing users to engage with both code and narrative within a familiar interface.

The paper describes a formative study where interviews with data analysts were conducted to identify workflows and challenges associated with actionable EDA and storytelling. These insights underpinned the development of the proposed design space. The system's usability and effectiveness were validated through a user study, which showed that participants preferred Jupybara over standalone LLM-based tools, citing superior response quality, usability, and transparency.

Jupybara aims to bridge the gap between traditional human expertise and the generative power of LLMs, demonstrating strong steerability, explainability, and reparability. The system supports analysts in generating and refining complex, diverse outputs, from raw data manipulations to sophisticated narratives, empowering them to communicate data insights more effectively.