An Exploration of Pattern Mining with ChatGPT

TL;DR

This exploratory work investigates using ChatGPT to mine patterns, introducing an eight-step collaborative process that blends human insight with AI capabilities. It demonstrates the approach by constructing a pattern language for integrating LLMs with data sources and tools, and argues for adding affordances of components as a new element in pattern descriptions. The study also discusses how prompts, iterative refinement, and pattern storytelling support the consolidation of practical patterns. While promising, it acknowledges limitations in generalizability and prompt dependence, outlining concrete future work to scale and refine the method. The work advances practical pattern mining with LLMs and provides a structured approach for pattern authors and tool designers.

Abstract

This paper takes an exploratory approach to examine the use of ChatGPT for pattern mining. It proposes an eight-step collaborative process that combines human insight with AI capabilities to extract patterns from known uses. The paper offers a practical demonstration of this process by creating a pattern language for integrating Large Language Models (LLMs) with data sources and tools. LLMs, such as ChatGPT, are a new class of AI models that have been trained on large amounts of text, and can create new content, including text, images, or video. The paper also argues for adding affordances of the underlying components as a new element of pattern descriptions. The primary audience of the paper includes pattern writers interested in pattern mining using LLMs.