RedNote-Vibe: A Dataset for Capturing Temporal Dynamics of AI-Generated Text in Social Media

Yudong Li, Yufei Sun, Yuhan Yao, Peiru Yang, Wanyue Li, Jiajun Zou, Yongfeng Huang, Linlin Shen

Keywords: AI-Generated Text, Social Media, User Engagement, RedNote(Xiaohongshu), Psycholinguistics

TL;DR: We provide the first longitudinal (5-years) dataset from RedNote platform for social media AIGT research.

Abstract: The proliferation of Large Language Models (LLMs) has led to widespread AI-Generated Text (AIGT) on social media platforms, creating new challenges for content authenticity. The identification of AIGT on social media platforms presents unique challenges due to engagement-driven content and temporal dynamics. To bridge this gap, we introduce a novel RedNote-Vibe dataset, collected from RedNote (Xiaohongshu), one of the most influential Chinese social media platforms. This dataset contains user posts and their parallel AIGT variants generated using diverse LLMs, spanning from before ChatGPT's release to the present. We further propose a detection method based on psycholinguistic principles, namely PsychoLinguistic AIGT Detection Framework (PLAD), which achieves SOTA performance compared to recent model-based methods and provides superior interpretability. Our analysis also reveals temporal trends of AI content adoption and engagement pattern differences between human and AI-generated content.

Primary Area: datasets and benchmarks

Submission Number: 4238