Botometer Pro

फ्रीमियम
द्वारा Observatory on Social Media | अपडेट किया गया 5 दिन पहले | Social
लोकप्रियता

9.1 / 10

लेटेंसी

1,080ms

सेवा का स्तर

100%

Health Check

N/A

सभी चर्चाओं पर वापस जाएं

Different results for the same Twitter account in short interval

Rapid account: Xixi Pa
XixiPa
3 वर्ष पहले

Hi,

I’m getting very different results for the same Twitter account in a very short interval.

By running the same query on @elonmusk I am getting the following results:

A more frequent result (± 90% of the time a similar result)
{‘cap’: {‘english’: 0.621687023966582, ‘universal’: 0.7828265255249504}, ‘displayscores’: {‘english’: {‘astroturf’: 0.4, ‘fakefollower’: 0.2, ‘financial’: 0.0, ‘other’: 1.3, ‘overall’: 0.8, ‘selfdeclared’: 0.1, ‘spammer’: 0.0}, ‘universal’: {‘astroturf’: 0.6, ‘fakefollower’: 0.4, ‘financial’: 0.0, ‘other’: 2.8, ‘overall’: 1.4, ‘selfdeclared’: 0.2, ‘spammer’: 0.0}}, ‘rawscores’: {‘english’: {‘astroturf’: 0.09, ‘fakefollower’: 0.04, ‘financial’: 0.0, ‘other’: 0.26, ‘overall’: 0.16, ‘selfdeclared’: 0.02, ‘spammer’: 0.0}, ‘universal’: {‘astroturf’: 0.11, ‘fakefollower’: 0.08, ‘financial’: 0.0, ‘other’: 0.55, ‘overall’: 0.29, ‘selfdeclared’: 0.05, ‘spammer’: 0.01}}, ‘user’: {‘majoritylang’: ‘en’, ‘userdata’: {‘idstr’: ‘44196397’, ‘screenname’: ‘elonmusk’}}}

A less frequent result (± 10% of the time a similar result):
{‘cap’: {‘english’: 0.7668769621923945, ‘universal’: 0.8053088638387352}, ‘displayscores’: {‘english’: {‘astroturf’: 0.4, ‘fakefollower’: 1.2, ‘financial’: 0.2, ‘other’: 2.7, ‘overall’: 1.4, ‘selfdeclared’: 1.5, ‘spammer’: 0.2}, ‘universal’: {‘astroturf’: 0.5, ‘fakefollower’: 0.6, ‘financial’: 0.4, ‘other’: 3.2, ‘overall’: 3.2, ‘selfdeclared’: 1.6, ‘spammer’: 0.1}}, ‘rawscores’: {‘english’: {‘astroturf’: 0.07, ‘fakefollower’: 0.25, ‘financial’: 0.05, ‘other’: 0.53, ‘overall’: 0.28, ‘selfdeclared’: 0.3, ‘spammer’: 0.03}, ‘universal’: {‘astroturf’: 0.1, ‘fakefollower’: 0.11, ‘financial’: 0.07, ‘other’: 0.65, 'overall’: 0.65, ‘selfdeclared’: 0.31, ‘spammer’: 0.02}}, ‘user’: {‘majoritylang’: ‘en’, ‘userdata’: {‘idstr’: ‘44196397’, ‘screenname’: ‘elonmusk’}}}

The probability of the account being a bot is more than doubled and passes the threshold score of 0.43 in the “universal” category (https://www.pewresearch.org/fact-tank/2018/04/19/qa-how-pew-research-center-identified-bots-on-twitter/), so it makes a big difference.

Any idea why this happens? Is this an expected behavior?

Thanks 😃

Rapid account: Xixi Pa
XixiPa Commented 3 वर्ष पहले

Thanks for the additional info on languages, very helpful.

Update on the variations: I picked a random Twitter user and I’m getting the same result each time so that was indeed the issue. A more typical account was a better pick for my tests.

Thanks a lot, your help is very appreciated

Rapid account: O So Me
OSoMe Commented 3 वर्ष पहले

Regarding language, the tweet object one obtains from Twitter’s API contains an field indicating the inferred language of the tweet. You can use this information to decide the language of the user and choose the proper bot score. You can, of course, use universal score in all cases.

– Botometer Team

Rapid account: Xixi Pa
XixiPa Commented 3 वर्ष पहले

Not only is this an helpful answer, it is also exceptionally fast!
@elonmusk is a very active account so it probably wasn’t the best test case, I’ll find a better one
Thanks a lot

Rapid account: O So Me
OSoMe Commented 3 वर्ष पहले

Botoemter use the most rencet 200 tweets of an account and other people’s tweets mentioning it to perform the evaluation. If a user is very active and posts many tweets in a short period of time and many other users mention this user, it could cause the bot score to fluctuate. Based on our own research, the variations for most of the users are not very significant. For the outliers, you can try to estimate the bot scores multiple times at different time and calculate the average.

Hope this is helpful.

–Botometer Team

Rapid account: Xixi Pa
XixiPa Commented 3 वर्ष पहले

I misplaced the bold elements but you get the idea, I cannot seem to edit my post
In short:

  • Situation 1: raw score - english -> 0.16 ; universal -> 0.29
  • Situation 2: raw score - english -> 0.28 ; universal -> 0.65

Are those the best values to take into account to identify a bot?

I try to identify bots in multiple languages. Is there an added-value to check if user is mostly speaking english (as majoritylang) and in that case use the english score, and the universal score if the user does not speak english? Or can I use the universal value in all cases?

Thanks

चर्चा में शामिल हों - नीचे टिप्पणी जोड़ें:

नई टिप्पणियाँ पोस्ट करने के लिए लॉग इन / साइनअप करें