Recommendation for Tabular Data Visualization Utilizing Machine Learning Applied to the Sales Domain

Effective data presentation and interpretation are essential for optimizing business operations, reducing costs, and enabling informed decision-making, ultimately fueling revenue growth and establishing a competitive advantage. This research addresses the challenge of recommending appropriate visualizations for tabular data based on user intentions expressed in natural language using Machine Learning for Visualization (ML4VIS). To tackle this challenge, we developed a comprehensive solution based on two complementary approaches. The first approach utilizes the BiDA4Sales model, which recommends a suitable visualization type by analyzing the tabular data and the user's natural language query (intention). The second approach employs a large language model (LLM) that further refines the visualization by suggesting which columns should be included, considering the recommended visualization type. The visualization-Type recommendation model achieved an F1-score of 90.24%, demonstrating high accuracy and reliability. In contrast, the column recommendation model achieved a correct recommendation percentage of 79.08%, indicating good precision in column selection for the previously recommended visualization type. Compared to general-purpose LLMs, our approach demonstrated superior performance in visualization type recommendation. These two approaches, when combined, enhance the flexibility and accuracy of visualization recommendations, making the process more efficient and user-friendly.