Twitter opens new timeline algorithm code – 4/3/2023 – Tech

Twitter published excerpts of its tweet recommendation algorithm on the open source platform GitHub last Friday (31). The code allows you to understand the workings of the technology behind the “For You” tab, with some programming knowledge.

According to a text published in platform engineering team blog, the initiative aims to increase the transparency of how the social network evaluates and delivers content. Twitter has been the target of criticism since Elon Musk bought the social network in October and implemented changes to the social network.

Before this movement for more transparency, Twitter had other snippets of its source code leaked on GitHub, by an account called “FreeSpeechEnthusiast”, an ironic reference to the unrestricted support for free speech that Musk claims to maintain. The social network called the US court and managed to remove the material from the air on the 27th.

Questioned via email by the report whether the leak had motivated the decision to open the source code, the platform automatically responded with an emoji in the form of feces, as has occurred with other questions from the press. Twitter no longer maintains an advisory team in Brazil since last year.

Musk wrote on Twitter that all source code for the social network should be made public in the future.

How tweet recommendation works

The Twitter engineering team’s paper describes a suggestion system compatible with what the “For You” tab delivers, says public interest technologist Lucas Lago of the Aaron Swartz Institute. “All the networks are scrambling to find the best resources for recommending content.”

The “For You” tab selects the tweets with which the user is most likely to interact, according to the material released by the platform. Posts made by accounts not followed by the user also appear on the timeline, if they pass the algorithm’s scrutiny.

According to Twitter, publications from people followed by the user are more likely to appear on the timeline. Posts by people outside your network, however, score better overall. “Today, the ‘For You’ tab consists, on average, of 50% in-network tweets and 50% out-of-network tweets.”

Like TikTok, the “For You” tab uses machine learning to curate the most popular content. To do this, use questions like:

• How likely are you to interact with this other user in the future?
• What are the communities on Twitter and which tweets are popular among these groups?
• What tweets have people I follow recently liked?
• Who likes tweets similar to the ones I like, and what else have they liked?

The technology is able to separate 1,500 potentially relevant publications from the answers to these questions. It then ranks those tweets with a neural network with 48 million parameters — ChatGPT, for comparison, has 175 billion parameters.

To arrive at this score, Twitter evaluates data from tweets, user and engagement information also with a neural network, based on artificial intelligence.

The selected and ranked tweets are then subject to filters, to veto content produced by blocked accounts, promote a variety of sources and avoid material from accounts that are not followed by at least one known person on the social network.

Users with a good relationship history also receive the best scores from the Twitter algorithm. Therefore, to have reach, it is recommended to avoid being blocked or having tweets reported.

According to Lago, the publication of Twitter code snippets shows that, today, the “For You” tab application does not privilege Elon Musk’s publications.

The suspicion arose after the newsletter specializing in social network coverage The Platformer, led by New York Times columnist Casey Newton, showed, based on internal documents from the social network, that the resource showed Musk’s tweets to 90% of the audience, after undergo modification at the request of the billionaire.

The operation of Twitter’s algorithm also showed that it uses labels to do audience analysis. For this, the technology separated tweets made by Musk, by accounts considered Democrats or by people verified before the launch of the Twitter Blue subscription package – which drew attention on social networks after the code was disclosed.

According to Lucas Lago, it doesn’t seem to be something with serious consequences. “Maybe Musk was trying to analyze the performance of his tweets?”