Read on Twitter
Out of the 170k CCP accounts Twitter deleted, they published the data for 23.8k accounts. which I analyzed.
My take: they are not bots, but actual human beings reluctantly taking part in a troll factory.
In this thread, I'll cover the specifics that really stand out.
THREAD:
My take: they are not bots, but actual human beings reluctantly taking part in a troll factory.
In this thread, I'll cover the specifics that really stand out.
THREAD:
1. Tweet timings: The tweets from the CCP accounts almost exclusively tweet during "work hours" by Beijing Time.
89% of the tweets were between 7 AM and 5 PM.
For the control group of 58k tweets from 32 accounts in HK & TW, that number was 37%!
89% of the tweets were between 7 AM and 5 PM.
For the control group of 58k tweets from 32 accounts in HK & TW, that number was 37%!
What it means: The average person is expected to tweet in their free time outside work hours. However, the CCP tweeters tweet almost exclusively during work-hours, which indicates that that is possibly their full-time job.
2. Devices used: 96% of the tweets in the CCP database are from a desktop/laptop.
This number is 18.2% for the control group.
This is reasonably expected because mobile app usage is much higher than desktop browser usage for social media.
This number is 18.2% for the control group.
This is reasonably expected because mobile app usage is much higher than desktop browser usage for social media.
In fact, 98.6% of the 23.8K accounts tweet exclusively using some sort of web client. Attached are the raw numbers FYI.
3. Retweets: As @JohnHemmings2 rightly pointed out, these accounts retweet a lot.
52.2% of all their tweets were retweets.
The corresponding number for the control group was less than 1%. This will need further study, but IMO it won't go beyond 20%.
52.2% of all their tweets were retweets.
The corresponding number for the control group was less than 1%. This will need further study, but IMO it won't go beyond 20%.
4. Locations: They appear to pick locations in Japan, Russia, Taiwan and the US for self-reporting. Attached are their top locations. Happy to share the raw data if needed.
Finally, something we have all seen that they often don't have a lot of followers. Here are the stats for those:
Mean follower count: 2.1
Mean following count: 2.8
Mean tweet count 14.7
Mean follower count: 2.1
Mean following count: 2.8
Mean tweet count 14.7
In summary, if I were to build a Twitter bot that identifies a CCP troll, I'd look at the % of tweets outside working hours, the client used and the % of retweets. 
