Visistant: A Conversational Chatbot for Natural Language to Visualizations With Gemini Large Language Models

Abstract

The goal of the data visualization field has always been to enable the production of visualizations directly from natural language text. This paper introduces Visistant, an innovative system designed to enhance the capabilities of advanced pre-trained language models, notably Google’s Gemini. Visistant facilitates the generation of interactive visualizations from tabular datasets through the use of Plotly and supports conversational interactions in a chatbot-styled manner by employing LangChain to maintain the memory of conversations. Visistant leverages Gemini’s code generation capabilities, demonstrating how prompt engineering can effectively lead to accurate end-to-end solutions. Our focus extended to prompt refinement, aiming to optimize the input token length of prompts. Through meticulous evaluation and comparison with existing models, Visistant exhibits superior performance in terms of both accuracy and efficiency. The results indicate that Visistant’s approach to converting natural language into detailed visualizations represents a significant advancement in the field of data visualization. The significance of this study lies in its potential to revolutionize data visualization by making it more accessible and user-friendly. By enabling users to generate complex visualizations through simple conversational interactions, Visistant democratizes data analysis, making it accessible to non-experts, which not only saves time but also reduces the barrier to entry for data-driven decision-making. This research addresses a critical gap in the literature by exploring the capabilities of Gemini LLMs for visualization generation, providing a novel solution that combines cutting-edge AI with practical usability.

Keywords

• Chatbot
• conversational interactions
• Gemini
• Langchain
• large language models
• prompt refining