A crawl of the social network Truth Social made in 2022: 845,060 posts (called Truths; 244,206 are ReTruths and 283,191 are replies) by 17,377 authors with their like, ReTruth and reply counts, 454,458 user records (21,955 scraped profiles and 432,503 placeholders created for users seen only as authors or in follow links), 4,002,115 follow links, 506,276 reply links and 10,508 quote links between users, and the hashtags (21,599), media files (184,884), external URLs (173,947) and user tags attached to truths, collected by Patrick Gerard, Nicholas Botzer and Tim Weninger with the accompanying paper at ICWSM 2023 (doi 10.1609/icwsm.v17i1.22211). Time columns are text because the data state no time zone.
The meme data of 'I Can Has Cheezburger? A Nonparanormal Approach to Combining Textual and Visual Information for Predicting and Generating Popular Meme Descriptions' (Wang and Wen, NAACL 2015): 252,435 captions of 110 memes (image macros such as Grumpy-Cat or Y-U-No) with the number of votes each received, 17,076 further meme descriptions from Cheezburger and 252,397 from Meme Generator used to build the paper's retrieval index, and the 117 meme images with the search phrase that Google's reverse image search returned for each. The upstream archive also holds the paper's feature files for the prediction and generation experiments (train.feat, test.feat, eval.feat, VisionFeatures.txt) and retrieval outputs; these are kept in the original files only.
M3 (Multi-platform, Multi-lingual, and Multimodal Meme), a benchmark for multimodal hate speech detection by Rui Xing, Qi Chai, Jie Ma, Jing Tao, Pinghui Wang, Shuming Zhang, Xinping Wang and Hao Wang: 2,455 memes, each an image with the text of the social media post it was shared in, collected from X (Twitter, 526 memes), 4chan (1,400) and Weibo (529). Each meme has the text read from its image (OCR), the post text, a label hate (1,318) or normal (1,137), the hate categories it falls in, and a short rationale. The hate categories are politics (348), race (334), violence (297), international relations (180), gender (172), religion (108), health status (106), public health (85); the 1,137 normal memes have the category none. One table of posts serves all three platforms; the column collection_member names the platform, so a single platform is a filter of the table.
The 4chan part of M3: 1,400 memes of M3 (image, OCR text, post text, label and rationale), 1,062 labeled hate and 338 normal. Hate categories: race (319), politics (251), violence (178), international relations (152), gender (144), health status (101), religion (88), public health (78).
The hateful memes of M3: 1,318 memes of M3 (image, OCR text, post text, label and rationale), 1,318 labeled hate and 0 normal. Hate categories: politics (348), race (334), violence (297), international relations (180), gender (172), religion (108), health status (106), public health (85).
The Weibo part of M3: 529 memes of M3 (image, OCR text, post text, label and rationale), 124 labeled hate and 405 normal. Hate categories: violence (86), politics (16), international relations (14), gender (12), race (10), religion (4), health status (2), public health (1).
The X (Twitter) part of M3: 526 memes of M3 (image, OCR text, post text, label and rationale), 132 labeled hate and 394 normal. Hate categories: politics (81), violence (33), gender (16), religion (16), international relations (14), public health (6), race (5), health status (3).
Truth Social
4chan
Twitter (X)
Weibo