DashBot: Insight-Driven Dashboard Generation Based on Deep Reinforcement Learning.

Area of Science:

• Computer Science
• Data Visualization
• Machine Learning

Background:

• Analytical dashboards are crucial for business intelligence but are complex to create, requiring specialized skills and tools.
• Existing deep learning approaches for dashboard generation are limited by the lack of large-scale, high-quality datasets.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) framework for automated analytical dashboard generation.
• To address the data scarcity issue in training machine learning models for visualization recommenders.

Main Methods:

• Utilized deep reinforcement learning with a focus on integrating established visualization knowledge.
• Constructed a training environment and reward system informed by visualization principles to guide agent learning.
• Designed a specialized agent network to imitate human exploration behaviors in dashboard creation.

Main Results:

• Demonstrated the effectiveness of the DRL model through comprehensive ablation studies.
• Validated the model's utility and performance via user studies, confirming its practical applicability.
• Showcased the model's ability to generate effective analytical dashboards without relying on pre-existing dashboard datasets.

Conclusions:

• The proposed DRL approach offers a viable solution for automated dashboard generation, reducing user burden.
• This work pioneers the development of machine learning-based visualization recommenders independent of large training datasets.
• Opens new avenues for creating intelligent visualization tools that leverage reinforcement learning and visualization expertise.