Are Large Language Models Ready for Multi-Turn Tabular Data Analysis?

Conversational Tabular Data Analysis, a collaboration between humans and machines, enables real-time data exploration for informed decision-making. The challenges and costs of collecting realistic conversational logs for tabular data analysis hinder comprehensive quantitative evaluation of Large Language Models (LLMs) in this task. To mitigate this issue, we introduce CoTA, a new benchmark to evaluate LLMs on conversational tabular data analysis. CoTA contains 1013 conversations, covering 4 practical scenarios: Normal, Action, Private, and Private Action. Notably, CoTA is constructed by an economical multi-agent environment, Decision Company, with few human efforts. This environment ensures efficiency and scalability of generating new conversational data. Our comprehensive study, conducted by data analysis experts, demonstrates that Decision Company is capable of producing diverse and high-quality data, laying the groundwork for efficient data annotation. We evaluate popular and advanced LLMs in CoTA, which highlights the challenges of conversational tabular data analysis. Furthermore, we propose Adaptive Conversation Reflection (ACR), a self-generated reflection strategy that guides LLMs to learn from successful histories. Experiments demonstrate that ACR can evolve LLMs into effective conversational data analysis agents, achieving a relative performance improvement of up to 35.14%.

What's this research about? Imagine chatting naturally with AI about your data: asking "Which products sold best?" then following up with "Show me the trend for those items", like talking to a human analyst. This research tackles the problem that we don't have good ways to test how well AI handles these natural data conversations. The team created CoTA, a benchmark with over 1,000 realistic conversations generated by "Decision Company," a virtual office where AI agents play different workplace roles and naturally discuss data. When they tested current AI agents, even advanced models struggled significantly with complex, multi-turn data conversations. Their solution, ACR (Adaptive Conversation Reflection), helps AI learn from successful past conversations, which is like a student reviewing their best work, leading to 35% better performance and bringing us closer to AI that can truly understand how we naturally want to explore data through conversation.