Table 9.

Multimodal Fake News datasets

Dataset	Year of release	Statistics	Domain	Contents	Labels	Collected from	Used in
Twitter 15 [144]	2015	361 (I) 7032 (F) 5008 (R)	Posts related to 11 events	Text, visual	2	Twitter	[4, 15, 26–28, 99]
Twitter 16 [89]	2016	413 (I) 9596 (F) 6225 (R)	Posts related to 17 events	Text, visual	2	Twitter	[25, 91, 111, 129]
Weibo [25]	2016	9528 (I) 4749 (F) 4779 (R)	Crawl the verfi ed false rumor posts from May, 2012 to Jan, 2016	Text, visual	2	Weibo (Non-rumor tweets are verifi ed by Xinhua News Agency, an authoritative news agency in China)	[4, 15, 25–28, 91, 91, 99]
PHEME [12]	2016	2672 (I) 1972 (F) 3830 (R)	9 different events, which include 5 cases of breaking news	Tweet, conversational threads	3	Twitter	[16, 92, 96, 99]
ALLData [100]	2018	20,015 (I) 11,941 (F) 8074 (R)	2016 US Presidential elections	The title, text, image, author and website	2	Fake and real news scraped from 240 websites and authoritative news websites, i.e., the New York Times, Washington Post, etc. respectively	[100, 110, 111]
FakeNewsNet [120]	2019	19,200 (I) 5367 (F) 17,222 (R)	Politics, Entertainment	Text, image url, conversational threads, location, and timestamp of engagement	2	Content is crawled from PolitiFact, GossipCop, E! online; For user engagements Twitter API is used	[94, 95, 97, 98]

Note: I—Total Number of Images, F—Number of Fake claims, R—Number of Real claims