Multimodal fake news dataset Weibo23

We propose a more challenging dataset known as Weibo23. By amalgamating all available fake news from the Weibo Management Community until March 2023 with existing samples from public datasets [1], we formed a comprehensive collection of fake news for Weibo23. Fabricated news articles were thoroughly examined and authenticated by certified experts. To facilitate the accurate differentiation between fake and real news, minimizing content-related disparities between them is imperative. Otherwise, the model tends to rely on specific content cues to identify fake news excessively. Therefore, for each instance of fake news, we employed the Baidu API to extract keywords and retrieved relevant news on Weibo based on the publication date and keywords. Subsequently, through manual scrutiny, all collected pertinent news was categorized into fake news, real news, and others (tweets about personal life, emotions, entertainment, etc.). The HANLP API was utilized to compute the similarity between samples within the same class and eliminate duplicates based on their similarity scores. Furthermore, samples collected before December 31, 2021 were partitioned into training and validation sets, while those obtained from January 1, 2022 onwards constituted the testing set.

[1] Q. Nan, J. Cao, Y. Zhu, Y. Wang, and J. Li, ‘‘Mdfend: Multi-domain fake news detection,’’ in Proceedings of the 30th ACM International Conference on Information & Knowledge Management, ser. CIKM ’21. New York, NY, USA: Association for Computing Machinery, 2021, p. 3343–3347. [Online]. Available: https://doi.org/10.1145/3459637.3482139