Scholarlink

Rumor or Fake news Detection (RFD) is of great importance in the era of We-media. Practical detection algorithms require training on feasible, timely, and social trend-aware datasets, which are quite limited, mainly due to the difficulties in collecting and labeling current rumors (especially Chinese rumors and fake news) on a large scale. In this paper, we make a pilot attempt to automatically acquire rumors more efficiently with the State Of The Art (SOTA) Chinese Large Language Models (LLM) of Kimi, Doubao, ZhipuQingyan, and DeepSeek etc. Due to the recent progress and widespread popularity of LLM, the semantic features of these generated fake news are nearly the same as the real ones. We also discover some tricks that allow these LLMs to generate some sensitive or restricted pieces of fake news. Some pioneering objective and subjective experiments illustrate that the proposed rumor dataset is quite challenging for the SOTA rumor detection algorithms.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A New Timely Dataset for Chinese Rumor Detection: Automatic Generation via SOTA Large Language Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shangrong Huang,
  • Junwei Wang,
  • Hua Deng,
  • Xiaowei Liu

摘要

Rumor or Fake news Detection (RFD) is of great importance in the era of We-media. Practical detection algorithms require training on feasible, timely, and social trend-aware datasets, which are quite limited, mainly due to the difficulties in collecting and labeling current rumors (especially Chinese rumors and fake news) on a large scale. In this paper, we make a pilot attempt to automatically acquire rumors more efficiently with the State Of The Art (SOTA) Chinese Large Language Models (LLM) of Kimi, Doubao, ZhipuQingyan, and DeepSeek etc. Due to the recent progress and widespread popularity of LLM, the semantic features of these generated fake news are nearly the same as the real ones. We also discover some tricks that allow these LLMs to generate some sensitive or restricted pieces of fake news. Some pioneering objective and subjective experiments illustrate that the proposed rumor dataset is quite challenging for the SOTA rumor detection algorithms.