📈HAIChart

📖Overview

HAIChart combines human insights and AI capabilities through reinforcement learning to enhance data visualization processes. This innovative system supports users with high-quality visualization recommendations and hints for interactive exploration, leveraging a two-part framework: the offline learning-to-rate component and the online multi-round recommendations.

• Offline Learning: Focuses on understanding and rating visualizations by training a neural network on a vast dataset of visualizations with their ratings.
• Online Recommendations: Uses a Monte Carlo Graph Search (MCGS)-powered agent to offer promising visualizations and hints, encouraging exploratory user interaction.

🎉News

[24/8/27] We presented our paper and poster at VLDB 2024, receiving widespread attention and positive feedback from researchers.

[24/6/30] Our paper HAIChart: Human and AI Paired Visualization System has been accepted by VLDB'24.

🎈Features

• Monte Carlo Graph Search: Auto-generates high-quality visualizations.
• Composite Reward Function: Effectively evaluates visualization quality.
• Visualization Hints Mechanism: Integrates feedback to fine-tune visualizations.

📁File Structure

• datasets contains the datasets used in this project, including VizML and KaggleBench.
• static houses web resources used by the project.
• templates includes the source code for the frontend interaction interface.
• tools folder contains components that organize this project, such as features.py, instance.py, etc.
• user_model folder holds the implementation code for the user model, including discriminator.py, agent.py, and more.
• mcgs.py is the implementation code for visualization generation and recommendation based on Monte Carlo Graph Search.
• haichart.py is the main script used to start the HAIChart web server. Running this script activates the backend server, enabling users to access the HAIChart web interface through their browser.

🛠️Environment Setup

To set up the environment for HAIChart, ensure you have Anaconda installed and then follow these steps:

1. Create and activate a new environment:

conda create -n haichart python=3.6.13
conda activate haichart

2. Install the necessary packages:

pip install -r requirements.txt

📊Demo Running Example

After setting up your environment, you can start exploring HAIChart with a basic demo. This example will guide you through the process of interacting with hints, selecting user constraints, and viewing recommended visualizations.

1. Start the HAIChart server:

python haichart.py

2. Open your web browser and navigate to the provided local server URL (typically http://localhost:8080).

3. In the HAIChart interface:

• Dataset Select and Display Area: Click 'Select Dataset' to select a dataset from your local system. Once uploaded, the dataset will be displayed in this area.
• User Preferences Area: This area dynamically generates preferences based on the hints you've interacted with previously. These preferences are derived from your responses to hints and help refine the visualization recommendations. Engage with this area to set or adjust your analysis parameters according to the insights you're seeking.
• Visualization Results Area: Based on your dataset and the constraints influenced by your interaction with hints, HAIChart will automatically recommend several visualizations. Explore these recommendations to uncover valuable insights.
• Visualization Hints Area: HAIChart provides hints to guide you through refining your visualizations or discovering new analytical angles. Interaction with these hints directly influences the constraints and recommendations, fostering a personalized exploration experience.

This example demonstrates the interactive workflow of HAIChart, emphasizing the importance of hints and user constraints in driving the discovery of insightful visualizations. Experiment with different datasets and engage with the hints to leverage the full analytical power of HAIChart.

📌Citation

@article{10.14778/3681954.3681992,
         author = {Xie, Yupeng and Luo, Yuyu and Li, Guoliang and Tang, Nan},
         title = {HAIChart: Human and AI Paired Visualization System},
         year = {2024},
         publisher = {VLDB Endowment},
         volume = {17},
         number = {11},
         issn = {2150-8097},
         url = {https://doi.org/10.14778/3681954.3681992},
         doi = {10.14778/3681954.3681992},
         journal = {Proc. VLDB Endow.},
         pages = {3178–3191},
         numpages = {14}
}

✉️Contact

If you have any questions, please contact:
[email protected]