Shulou(Shulou.com)11/24 Report--

According to the news on April 7, last week Twitter fulfilled a long-standing promise made by Elon Musk by posting some of the underlying code of the recommendation algorithm on GitHub. However, the researchers say that the content is not exhaustive, and Twitter omits important details such as "how the algorithm works," making it difficult for us to understand the inner workings of the Twitter platform.

The following is the translation:

When Musk first offered to take over Twitter, he claimed that the first thing he did was the "open source" Twitter algorithm. Last week, he finally fulfilled his promise when Twitter released some of the underlying code on the platform about the "recommend to you" feature on GitHub.

People soon began to dig deep into the code, and it wasn't long before Musk's tweets were marked with special categories. Twitter engineers rushed to explain that this was to facilitate "tracking statistics", a claim confirmed by other analyses. Although Twitter removed the code from GitHub a few hours later, there is still speculation that Twitter engineers have given preferential treatment to Musk's tweets and may have taken artificial steps to increase the exposure of his tweets.

Since then, Twitter has not released any more code or explained the internal workings of its platform's algorithms. Those who want to explore the inner workings of the platform through Twitter's public code may be disappointed. Engineers who have studied the algorithms say this is because the code released by Twitter omits key details such as how the "algorithm" works.

According to Sol Messing, an associate professor of social media at New York University and a former Twitter employee, Twitter's public content makes "a lot of edits" of Twitter algorithms, and the first problem is that it does not cover all the systems that play a role in Twitter recommendations.

Twitter said it did not disclose the code, trust and security systems used to handle ads to prevent bad actors from using the code for malicious operations. In addition, Twitter chose not to publish the underlying model used to train the algorithm, explaining in a blog post that this was to "ensure the security and privacy of users". Maxine believes the decision is even more important. "the most critical part of the algorithm, the driver model, is still not open source," he said. "so the core of the algorithm is still confusing."

Musk's motivation for opening up the Twitter algorithm in the first place seems to be his concern that Twitter is using the algorithm to suppress freedom of speech. "I think Twitter should open source algorithms, and any changes to the content of Twitter, whether recommended or not, should be clearly presented," he said in a TED speech last April. "in this way, anyone can see what measures Twitter has taken, and there will be no algorithmic or manual behind-the-scenes operations." The speech was made shortly after Musk confirmed his acquisition of Twitter.

The code released on Twitter did not provide enough information to tell people whether the platform was potentially biased or whether there was the kind of "behind-the-scenes manipulation" that Mr Musk wanted to disclose. "it's just a matter of transparency," Maxine said. "but it doesn't really help people understand how the algorithm works, nor does it really give people an insight into why some people's tweets go down while others go up."

In addition, Messin also pointed out that Twitter's recent policy on adjusting its application programming interface (API) has basically cut off the vast majority of researchers' access to a large amount of data on the Twitter platform. Without proper API access, it is impossible for researchers to study the new details of Twitter algorithms on their own. "while Twitter publishes the code, this makes it more difficult for researchers to analyze the code in depth," Maxine wrote in his analysis. "

Last year, shortly after Mr Musk first discussed his plans for "open source" Twitter algorithms, Alex Hanna, research director of the distributed artificial Intelligence Institute (DAIR), also mentioned the importance of researchers reviewing code. Like Maxine, she believes that posting code on GitHub alone is not enough to improve the transparency of the way Twitter works.

"if you're really interested in things like public scrutiny of Twitter algorithms, you have to take a variety of approaches," Hannah said. "

However, some of the code exposed by Twitter on GitHub does reveal something about the Twitter algorithm. Data scientist Jeff Allen (Jeff Allen) has unearthed a document that reveals how Twitter uses "formulas" to prioritize different types of user participation, according to Messin. "if we take it literally, one like on Twitter is equivalent to half a tweet," Maxine wrote. "one reply is equivalent to 27 retweets, while one reply plus the tweet author's response is equivalent to 75 retweets."

While this reveals some information to some extent, the description of how the Twitter platform works is incomplete. "without actual data, this kind of analysis doesn't make much sense," Maxine said. "now Musk makes it ridiculously expensive for scholars to obtain data. if they really want to study the problem now, it will basically cost a huge amount of $500000 a year (currently about 3.44 million yuan) to get a lot of meaningful data."

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.