HAIChart: Human and AI Paired Visualization System

Yupeng Xie,Yuyu Luo,Guoliang Li,Nan Tang

2024-06-17

Abstract:The growing importance of data visualization in business intelligence and data science emphasizes the need for tools that can efficiently generate meaningful visualizations from large datasets. Existing tools fall into two main categories: human-powered tools (e.g., Tableau and PowerBI), which require intensive expert involvement, and AI-powered automated tools (e.g., Draco and Table2Charts), which often fall short of guessing specific user needs. In this paper, we aim to achieve the best of both worlds. Our key idea is to initially auto-generate a set of high-quality visualizations to minimize manual effort, then refine this process iteratively with user feedback to more closely align with their needs. To this end, we present HAIChart, a reinforcement learning-based framework designed to iteratively recommend good visualizations for a given dataset by incorporating user feedback. Specifically, we propose a Monte Carlo Graph Search-based visualization generation algorithm paired with a composite reward function to efficiently explore the visualization space and automatically generate good visualizations. We devise a visualization hints mechanism to actively incorporate user feedback, thus progressively refining the visualization generation module. We further prove that the top-k visualization hints selection problem is NP-hard and design an efficient algorithm. We conduct both quantitative evaluations and user studies, showing that HAIChart significantly outperforms state-of-the-art human-powered tools (21% better at Recall and 1.8 times faster) and AI-powered automatic tools (25.1% and 14.9% better in terms of Hit@3 and R10@30, respectively).

Databases,Artificial Intelligence

What problem does this paper attempt to address?

### The Problem the Paper Attempts to Solve

This paper aims to address a core issue in the field of data visualization: how to combine the knowledge of human experts with the capabilities of artificial intelligence (AI) to generate high-quality data visualizations. Specifically, existing data visualization tools can be categorized into two types:

1. **Human-driven tools** (such as Tableau, PowerBI), which require significant expert involvement. While the results are of high quality, the process is inefficient and prone to errors.
2. **AI-driven automated tools** (such as Draco, Table2Charts), which can automatically generate visualizations but lack the ability to understand specific user needs.

To overcome the limitations of these two approaches, this paper proposes the HAIChart system, which combines human feedback with AI capabilities to achieve its goals through the following steps:
1. **Automatic recommendation of initial high-quality visualizations**: Using a reinforcement learning framework, HAIChart first generates a set of high-quality charts to reduce the user's initial workload.
2. **Iterative optimization**: Continuously adjusts and optimizes the recommended charts based on user feedback to better meet specific user needs.

### Specific Contributions

1. **Problem Definition**: Formally defines the problem of human-AI collaborative data visualization.
2. **HAIChart System**: Proposes a reinforcement learning-based method that combines human feedback and AI capabilities to generate higher quality visualizations.
3. **Evaluation Mechanism Design**: Designs a composite reward function to evaluate the quality of the charts, guiding the generation process.
4. **Prompt Selection Algorithm**: Designs a visualization prompt selection algorithm for collecting user feedback and proves its NP-hard problem characteristics.
5. **Experimental Validation**: Demonstrates through quantitative evaluation and user studies that HAIChart outperforms existing human-driven and AI-driven tools in terms of accuracy and efficiency.