What is the mechanism of the algorithm that determines the display / non-display of community notes of X (old Twitter)

The mechanism of X (formerly Twitter)'s

community note is open source and open to the public, so that anyone can check the contents. Vitalik Buterin, the creator of Ethereum, has analyzed such community notes on his blog.

What do I think about Community Notes?

Since completing the acquisition of Twitter (now X) in 2022, Elon Musk has radically changed many aspects of the company, including staffing, content moderation, and business model. There are various opinions about whether these changes were good or bad, but one of the changes, 'Community Note', is widely accepted across various positions. .

Community notes are tools for adding information to tweets, and are used to counter hoaxes and provide supplementary explanations. For example, as shown in the figure below, ``Elon Musk may threaten free speech on Twitter by literally allowing people to speak freely. It is an image created by , and it is not broadcasted by CNN, ”the fact is supplemented.

The conditions for participating in the community note are as follows.

- Hasn't broken any rules recently
・More than 6 months have passed since the account was created
・Phone number verified

Anyone who meets the above conditions can participate in the Community Note. It is not possible to create a community note from the beginning, and it is necessary to first evaluate existing notes and increase your own score. If the vote for an existing note matches the final result of that note, your score will improve, and if you earn a score above a certain level, you will be granted the right to create a new community note.

There are basically three evaluations for community notes: ``useful'', ``partially useful'', and ``not useful'', but there are cases where different tags are assigned internally. Community notes with a total evaluation result exceeding 0.4 will be displayed to general users. However, this tally is not a simple average value, but an algorithm that increases the evaluation when receiving positive evaluations from people in various positions.

Consider a matrix with users vertically, community notes horizontally, and the results of how users rated each community note. In this matrix, the 'i' row is 'Rating from i'th user', the 'j' column is 'Rating to j'th community note', and the 'i, j' address is 'Rating from i'th user The rating given to the jth community note' is displayed. The matrix is mostly '0' because each user has a small number of notes to rate.

For the values in this matrix, create a model like this:

A description of each variable in the model follows.

u represents user, n represents note, and M_un on the left side is the predicted value of the evaluation result of address 'u, n' in the matrix.

A 'Public Mood' parameter that describes how ratings users generally rate.

A parameter that represents the 'friendliness' of the u-th user, representing how much they generally give.

A parameter that represents the 'usefulness' of the community note . Ultimately, you'll want to find a numerical value for this parameter.

・f_u and f_n
Respectively, f_u is the user's 'polarity' and f_n is the parameter representing the community note's 'polarity'. Similar polarity values are automatically assigned to groups of users and community notes that exhibit similar tendencies, and so far, negative values for politically 'left' user notes, A positive value is assigned to politically 'right-wing' user notes. Note that these distinctions are automatically discovered by the algorithm, and the source code does not embed left-wing or right-wing concepts.

The community note algorithm uses the steepest descent method to find each parameter. Ultimately, 'i_n' becomes the evaluation of the community note, and community notes with 'i_n' of 0.4 or more are displayed to general users.

Mr. Britten, who conducted the analysis, highly appreciates the introduction of the concept of 'polarity'. By using the polarity parameter, it is possible to absorb ``characteristics that are preferred only by some users'' and measure only the characteristics that cause community notes to be widely preferred.

For example, the community note whose polarity is near the value of '-0.8' is as follows. You can see that community notes with a large swing in polarity have strong partisan content.

community notes polarity
Anti-trans rhetoric has been amplified by some conservative Colorado lawmakers, including US Rep. Lauren Boebert, who narrowly won re-election in Colorado's GOP-leaning 3rd Congressional District, which does not include Colorado Springs. https://coloradosun.com/ 2022/11/20/colorado-springs-club-q-lgbtq-trans/
(The anti-trans rhetoric has been amplified by some conservative Colorado legislators, such as Rep. Lauren Boebert, who narrowly won re-election in the Republican-leaning Colorado Third District, which does not include Colorado Springs. there is.)
President Trump explicitly undermined American faith in election results in the months leading up to the 2020 election. -before-they-attacked Enforcing Twitter's Terms of Service is not election interference.
(President Trump blatantly undermined Americans' confidence in election results in the months leading up to the 2020 election. Enforcing Twitter's terms of service is not election interference.)
The 2020 election was conducted in a free and fair manner. https://www.npr.org/2021/12/23/1065277246/trump-big-lie-jan-6-election
(The 2020 elections were held in a free and fair manner.)

In addition, the community note when the polarity swings to the vicinity of '+0.8' and the other side has the following contents.
community notes polarity
As of 2021 data, 64% of 'Black or African American' children lived in single-parent families. https://datacenter.aecf.org/data/tables/107-children-in-single-parent-families-by- race-and-ethnicity
(In 2021 data, 64% of “black or African American” children lived in single-parent households.)
Contrary to Rolling Stones push to claim child trafficking is 'a Qanon adjacent conspiracy,' child trafficking is a real and huge issue that this movie accurately depicts. Operation Underground Railroad works with multinational agencies to combat this issue. https://ourrescue.org /
(Contrary to what the Rolling Stones claim child trafficking is a 'QAnon-adjacent conspiracy', child trafficking is a real, big problem, and this movie accurately portrays it. 'Operation Underground Railroad' is working with multinational agencies to combat this problem.)
Example pages from these LGBTQ+ children's books being banned can be seen here: https://i.imgur.com/8SY6cEx.png These books are obscene, which is not protected by the US constitution as free speech. https://www. justice.gov/criminal-ceos/obscenity 'Federal law strictly prohibits the distribution of obscene matter to minors.
(Example pages of banned LGBTQ+ children's books can be found here: https://i.imgur.com/8SY6cEx.png These books are obscene and protected by the U.S. Constitution as free speech. No. Federal law strictly prohibits the distribution of obscene material to minors.

Britten also examined how much the introduction of 'polarity' would 'improve' the results over a simple average. The average rating of the 'good' group, which receives a '+2' rating from users with the same polarity and a '0' rating from users with the opposite polarity, is about 0.30, and a '+4' rating from users with the same polarity. The average rating of the 'biased good' group, which received a '-2' rating from users of the opposite polarity in the evaluation, was about 0.22. Even if the average rating is the same, you can see that community notes with a strong polarity are less likely to be displayed.
group average rating
biased good

The core part of the algorithm is shown above, but there are many other systems at work. The entire system is open to the public on X's site , and anyone can check it. In addition, all community notes and evaluation data are also published , so please check whether the algorithm is really working as explained and whether the management is switching between public and private without permission. I can.

Mr. Britten said that while community notes cannot introduce another mechanism if they don't like community notes, they are great because they guarantee a platform that doesn't do centralized operations or engage in cheating. They say they provide value. It's also a valuable example of something designed to bridge the two rather than perpetuate conflicting structures, and he concludes by saying that he looks forward to seeing the development of community notes and similar algorithms in the future. .

in Software,   Web Service, Posted by log1d_ts