Data Analytics with Large Language Models (LLM): A Novel Prompting Framework

This study presents a novel framework for conducting data analytics using Large Language Models (LLMs). The proposed framework suggests the construction of prompts and interaction patterns using four fundamental constructs: meta-specifications, specifications, instructions, and prompting patterns. The framework can guide and assist data engineers, analysts, and even non-technical domain experts by providing these four constructs as palettes of options. The LLM can then suggest analytics designs, conduct the analysis, provide posterior interpretations and insights, and produce other outputs, such as code or packaged software. The presented novel framework covers an immense space of possibilities through numerous combinations of selected meta-specifications, specifications, instructions, and prompting patterns. The primary theoretical contribution of this research is that it proposes a theoretical foundation and frame of reference for conducting data analytics using LLM. The primary practical contribution is that LLMs can now be employed much more systematically and extensively than before in designing and conducting data analytics. This opens a new world of applications powered by a countless combination of the four constructs across practically all fields of science, technology, and business, where LLMs can be used to guide, conduct, and interpret the results of data analytics.