Abstract
Agent-based conversational systems powered by large language models (LLMs) are emerging as promising tools for facilitating several tasks, from software development, e-commerce and customer interaction, and scientific discovery, to natural language to visualization (NL2VIS) tasks. These systems enable users to generate charts and extract insights from structured data through natural language interaction, significantly lowering technical barriers for non-specialist users. However, a comprehensive synthesis of recent developments in this rapidly evolving area has been lacking. In this work, we conduct a systematic literature review guided by the PRISMA methodology, focusing on publications that investigate the creation of LLM-based agents- systems for visual analytics. There is a notable acceleration in research activity in this area following the widespread deployment of LLMs, particularly after 2022. Findings show that recent systems increasingly adopt multi-agent reasoning, context-aware prompting, and code generation capabilities. Whereas these tools hold great potential for democratizing data analysis, current limitations include prompt fragility, model hallucinations, and inconsistent performance across domains. This review provides a comprehensive overview of NL2VIS agent-based systems and offers a foundation for future research in this growing field.
Funding Information
  • Funda??o para a Ci?ncia e a Tecnologia (2024.07501.IACDC)
  • Funda??o para a Ci?ncia e a Tecnologia (FCT UIDB/04152/2020)
  • NextGenerationEU (C645192610-00000060)

This publication has 60 references indexed in Scilit: