Reproduce “Early ChatGPT User Portrait through the Lens of Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuyang Deng, Ni Zhao, Xin Huang
An exploratory analysis of 87,957 publicly shared ChatGPT conversations characterizes early user behavior through interaction length, language, sentiment, and topic patterns. The study finds that conversations are typically short, English dominates across 45 detected languages, programming and data-related tasks are prevalent, and topic frequencies shift with contemporaneous events, while also highlighting the limits of self-selected ShareGPT data.
Key insights
- Interaction is usually brief: the median conversation contains six turns, 27.7% end within two turns, and more than 65.5% finish within ten turns, although a small number extend to 998 turns.
- English accounts for over 70% of conversations, while 45 languages are represented; sentiment is generally positive or neutral in user messages and predominantly positive in ChatGPT responses.
- Programming, coding, data analysis, business, finance, writing, and interpersonal communication recur as major use areas, while changing topic distributions track events such as the Russia–Ukraine conflict and international sports.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.