If you wanted to learn more about Sina Weibo’s censorship patterns, today is your lucky day. A group of computer scientists from Bowdoin College, Rice College, and the University of New Mexico have, along with an independent researcher, released the results of an academic study of Sina Weibo’s censorship practices. The study, which we came across via MIT Technology Review, used “architecture [that could] detect post deletions within one minute of the deletion event,” giving the researchers perhaps the most precise look yet into how quickly Sina’s content team takes down sensitive Weibo posts. The results? Sina is pretty darn fast:
We found that deletions happen most heavily in the ﬁrst hour after a post has been submitted. Focusing on original posts, not reposts/retweets, we observed that nearly 30% of the total deletion events occur within 5-30 minutes. Nearly 90% of the deletions happen within the ﬁrst 24 hours.
So Sina’s censors are pretty fast. But what, exactly, are they deleting? Researchers used a variety of analytical tools to look at what content was most quickly deleted, and found that:
The topics where mass removal happens the fastest are those that combine events that are hot topics in Weibo as a whole (e.g., the Beijing rainstorms or a sex scandal) with themes common to sensitive posts (e.g., Beijing, government, China, and policeman).
Researchers also found that, unsurprisingly, users with more total deleted posts tended to get their posts deleted more quickly than other users, suggesting that Sina’s content team was watching their accounts more carefully. The following chart from the study shows the downward trend in post lifetime as a user’s number of total deleted posts increases:
Of course, it’s not all humans doing the deleting. In fact, by the study’s estimations, for an all-human team to censor Weibo, 4,200 team members would be required, assuming each team member could read at the blazing rate of 50 posts per minute. The study points out that as a result of that, weibo’s censorship system has become an incredibly complex system, employing both human and software censors, employing multiple blocked keyword lists that trigger different censorship responses, search filtration systems, and more. (Of course, none of that should come as much of a surprise to longtime weibo users, who have likely experienced many of the different types of censorship on Sina Weibo firsthand).
If you’re really interested in Weibo censorship, the full paper is worth a read, and although it’s a bit dry and quite technical in places, the good news is that it’s only ten pages long.
(via MIT Technology Review)