Natural Language Interface for Data Visualization: Harnessing the Power of Large Language Models
Buchepalli Praneeth, Ashutosh Kumar Singh, Jyothika K Raju, Mohana, B Suma · 2024
Democratizing data analysis is the need of the hour, empowering a wider range of users within corporations to engage with data and uncover valuable insights. Because of the inherent ambiguity of natural language and the frequent lack of quality and clarity in user queries, implementing functioning Natural Language Interfaces (NLIs) has proven to be a challenging task. These problems make it difficult for existing language models to understand user intent effectively. This study introduces a novel data visualization tool that leverages the power of Large Language Models (LLMs) to empower corporations and research institutions in extracting meaningful insights from vast datasets. LLMs interpret user input and generate goals, which LIDA, a library developed by Microsoft, uses to create a wide range of visualizations with libraries like matplotlib, seaborn, and plotly, based on these goals and user instructions. This system bridges the gap between Natural Language Interfaces(NLIs) and data visualization by enabling users to submit prompts in natural language. Users can formulate questions in natural language, eliminating the need for complex code or data manipulation expertise. This work compares and contrasts the performance of open source models Mistral 7B instruct-v3, LLaMa3 8B, and proprietary model, gpt 3.5-turbo across several case studies as the graphs generated are dependent completely on the interpretation power of LLMs.