Tech

‘We Are Very Free’: How China Spreads Its Propaganda Version of Life in Xinjiang


Methodology

The data analyzed for this article includes videos, metadata and social media posts collected from Twitter, YouTube and Pomegranate Cloud between Feb. 18 and June 2. We downloaded more than 5,000 videos posted to these platforms between Jan. 23 (the date of the first campaign video following the State Department’s Jan. 19 declaration of genocide in Xinjiang) and May 31. On Pomegranate Cloud, we collected clips targeting Mr. Pompeo by searching for posts mentioning him after Jan. 19 that contained video. We collected videos denying forced labor in Xinjiang’s cotton industry from a section dedicated to them in the app.

On Twitter and YouTube, the campaign videos were collected from what we call “warehouse accounts,” those whose videos were shared by a network of more than 300 coordinated Twitter accounts. This network appeared to work in coordination to like and retweet content that supported Chinese government policies, such as the campaign videos, as well as news articles and editorials from state media. To define the network, we manually identified a small group of accounts and their indicators of automation, specifically posts containing identical content followed by strings of random characters. We then identified additional network accounts by programmatically searching for other accounts that boosted the same content and had the same indicators.

We cataloged more than 3,000 unique campaign videos out of the more than 5,000 collected. To pinpoint duplicates among videos containing various compression rates, visual artifacts and subtitle languages, we calculated a fingerprint for each video by running a sample of its frames through the Google Cloud Vision image labeler. We determined videos with similar fingerprints and durations to be duplicates. We manually sampled and reviewed the results from this process to minimize false positives and false negatives.

To identify non-campaign videos on YouTube and Twitter, we first obtained their subtitles by using optical character recognition on frames taken from them at regular intervals. We considered videos that did not mention Mr. Pompeo or cotton to be non-campaign videos. Videos on Twitter and YouTube always had subtitles in English and Chinese. Videos from Pomegranate Cloud had subtitles in Chinese only when Uyghur was spoken. We considered all videos collected from Pomegranate Cloud to be campaign videos, but did not review each manually. We also used the subtitles of the YouTube and Twitter videos to analyze their content.

Most Related Links :
newsbinding Governmental News Finance News

Source link

Back to top button