DOCUMENTATION
DOCUMENTATION
Embedded Files
Dataset description
Dataset description
A total of 2,000 URLs were collected from 272 channels: 1,190 URLs from YouTube, 595 URLs from TikTok, 111 URLs from Reddit, and 100 URLs from BiliBili, following the described procedure.
If a platform was chosen, to identify deepfake videos, we included only those videos that were clearly labeled as deepfakes in their titles. We conducted searches using the top four countries (USA, Korea, China, and Russia) producing deepfake content, using keywords such as "Deepfake", "딥페이크","深度偽造","дипфейк".
Dataset analysis
Dataset analysis
Purpose of deepfake generation.
Geographic distribution of publisher.
Proportion of deepfake applications.
Growth across platforms.
Distribution of victim and driver gender across ethnicities in our dataset, with the x-axis representing ethnicities and the y-axis representing the total number of cases/counts.
Distribution of the average number of likes per video over different victim ethnicities.
Temporal changes in video comment counts and their corresponding median sentiment scores, with red solid vertical lines indicating significant events.
Sentiment score distribution. The higher the score, the more positive viewers' attitudes are towards the videos.
Page updated
Google Sites
Report abuse