notesum.ai

Published at December 10

Towards Automated Cross-domain Exploratory Data Analysis through Large Language Models

cs.DB

cs.AI

H.2.8

Released Date: December 10, 2024

Authors: Jun-Peng Zhu¹, Boyan Niu², Peng cai³, Zheming Ni², Jianwei Wan², Kai Xu², Jiajun Huang², Shengbo Ma², Bing Wang², Xuan Zhou³, Guanglei Bao², Donghui Zhang², Liu Tang², Qi Liu²

Aff.: ¹East China Normal University & PingCAP, China; ²PingCAP, China; ³East China Normal University, China

Arxiv: http://arxiv.org/pdf/2412.07214v1

Methods	Test EX (%)
Methods	easy	medium	hard	extra	all
TiSQL + Claude-3 Opus	91.1	86.8	80.8	72.5	84.1
TiSQL + Claude-3 Haiku	84.5	84.0	61.6	55.7	74.6
TiSQL + Claude-3 Sonnet	90.0	83.8	67.0	65.3	78.4
TiSQL + GPT-4o	91.9	86.2	79.9	73.1	83.9
TiSQL + GPT-4o mini	91.3	68.1	73.0	59.4	72.8
TiSQL + GPT-4	89.8	89.0	83.6	78.7	86.3