Benchmark du modèle Jev dans la classification de tweets
How the Twitter/X Algorithm Works in 2026 (Source Code)
Twitter published its ranking weights on GitHub. A reply is worth 150x a like. Free accounts get 10x less reach. Full scoring system explained.
Social Media Feed Ranking Algorithms: Guide to Field Experiments
IC2S2'26 Tutorial | Social Media Feed Ranking Algorithms: Guide to Field Experiments
Community Archive
An open, searchable archive of tweets that members of the community have chosen to make public. Browse profiles, search conversations, and build on the data.
AI Fact-Checking in the Wild: A Field Evaluation of LLM-Written Community Notes on X
Large language models (LLMs) show promising capabilities for fact-checking, yet prior work evaluates them only in controlled offline settings using benchmarks or crowdworker judgments. Success in real-world fact-checking depends also on how content is judged within a live platform environment. We present the first field evaluation of LLM fact-checking deployed on a live social media platform, testing performance directly through X Community Notes’ “AI writer” feature over a three-month period. Our LLM writer, a multi-step pipeline that handles multimodal content, conducts web and platform-native search, and writes contextual notes, was deployed to write 1,614 notes on 1,597 tweets and compared against 1,332 human-written notes on the same tweets using 108,169 ratings from 42,521 raters. Direct comparison of note-level platform outcomes is complicated by differences in submission timing and exposure between LLM and human notes; we therefore pursue two analysis strategies: a rating-level analysis modeling individual rater evaluations, and a note-level analysis that relies on common raters who rated all notes on the same post. Rating-level analysis shows that LLM notes receive more positive ratings than human notes across raters with different political viewpoints, and note-level analysis shows LLM notes achieve significantly higher helpfulness scores among common raters. Rater-provided tags suggest that people consider LLM notes to use more neutral language and cite better sources. These findings provide field evidence that LLMs can contribute broadly helpful fact-checking notes at scale, while showing that their evaluation is shaped by platform dynamics absent from controlled offline settings.
How is science clicked on Twitter? Click metrics for Bitly short links to scientific publications
Abstract To provide some context for the potential engagement behavior of Twitter users around science, this article investigates how Bitly short links to scientific publications embedded in scholarly Twitter mentions are clicked on Twitter. Based on the click metrics of over 1.1 million Bitly short links referring to Web of Science (WoS) publications, our results show that around 49.5% of them were not clicked by Twitter users. For those Bitly short links with clicks from Twitter, the majority of their Twitter clicks accumulated within a short period of time after they were first tweeted. Bitly short links to the publications in the field of Social Sciences and Humanities tend to attract more clicks from Twitter over other subject fields. This article also assesses the extent to which Twitter clicks are correlated with some other impact indicators. Twitter clicks are weakly correlated with scholarly impact indicators (WoS citations and Mendeley readers), but moderately correlated to other Twitter engagement indicators (total retweets and total likes). In light of these results, we highlight the importance of paying more attention to the click metrics of URLs in scholarly Twitter mentions, to improve our understanding about the more effective dissemination and reception of science information on Twitter.
Note from 8 February 2024
There are now 68 Twitter archives mapped in `tweetback-canonical`, which means links to the original tweets in other tweetback archives are transformed…
Infiltré dans la fachosphère de Twitter/X
À quoi ressemble X pour un sympathisant d'extrême droite ? Les algorithmes de recommandation délivrant des contenus radicalement différents en fonction des affinités de l'utilisateur, que voient les …
Evaluating Twitter’s algorithmic amplification of low-credibility content: an observational study
Artificial intelligence (AI)-powered recommender systems play a crucial role in determining the content that users are exposed to on social media platforms. However, the behavioural patterns of these systems are often opaque, complicating the evaluation of their impact on the dissemination and consumption of disinformation and misinformation. To begin addressing this evidence gap, this study presents a measurement approach that uses observed digital traces to infer the status of algorithmic amplification of low-credibility content on Twitter over a 14-day period in January 2023. Using an original dataset of ≈ 2.7 million posts on COVID-19 and climate change published on the platform, this study identifies tweets sharing information from low-credibility domains, and uses a bootstrapping model with two stratifications, a tweet’s engagement level and a user’s followers level, to compare any differences in impressions generated between low-credibility and high-credibility samples. Additional stratification variables of toxicity, political bias, and verified status are also examined. This analysis provides valuable observational evidence on whether the Twitter algorithm favours the visibility of low-credibility content, with results indicating that, on aggregate, tweets containing low-credibility URL domains perform better than tweets that do not across both datasets. However, this effect is largely attributable to a difference in high-engagement, high-followers tweets, which are very impactful in terms of impressions generation, and are more likely receive amplified visibility when containing low-credibility content. Furthermore, high toxicity tweets and those with right-leaning bias see heightened amplification, as do low-credibility tweets from verified accounts. Ultimately, this suggests that Twitter’s recommender system may have facilitated the diffusion of false content by amplifying the visibility of low-credibility content with high-engagement generated by very influential users.
Evaluating Twitter’s algorithmic amplification of low-credibility content: an observational study
Artificial intelligence (AI)-powered recommender systems play a crucial role in determining the content that users are exposed to on social media platforms. However, the behavioural patterns of these systems are often opaque, complicating the evaluation of their impact on the dissemination and consumption of disinformation and misinformation. To begin addressing this evidence gap, this study presents a measurement approach that uses observed digital traces to infer the status of algorithmic amplification of low-credibility content on Twitter over a 14-day period in January 2023. Using an original dataset of ≈ 2.7 million posts on COVID-19 and climate change published on the platform, this study identifies tweets sharing information from low-credibility domains, and uses a bootstrapping model with two stratifications, a tweet’s engagement level and a user’s followers level, to compare any differences in impressions generated between low-credibility and high-credibility samples. Additional stratification variables of toxicity, political bias, and verified status are also examined. This analysis provides valuable observational evidence on whether the Twitter algorithm favours the visibility of low-credibility content, with results indicating that, on aggregate, tweets containing low-credibility URL domains perform better than tweets that do not across both datasets. However, this effect is largely attributable to a difference in high-engagement, high-followers tweets, which are very impactful in terms of impressions generation, and are more likely receive amplified visibility when containing low-credibility content. Furthermore, high toxicity tweets and those with right-leaning bias see heightened amplification, as do low-credibility tweets from verified accounts. Ultimately, this suggests that Twitter’s recommender system may have facilitated the diffusion of false content by amplifying the visibility of low-credibility content with high-engagement generated by very influential users.
Yes, Elon Musk created a special system for showing you all his tweets first
After his Super Bowl tweet did worse numbers than President Biden’s, Twitter’s CEO ordered major changes to the algorithm.
How Twitter’s Algorithm Works: A High-level Breakdown
On March 31st, 2023, Twitter made an unprecedented move by releasing the source code for their algorithm. The move is unprecedented because…
The Future of Twitter: Q&A with Jack Dorsey
An in-depth conversation on 10 years of tweeting, the importance of live media, and “the puddle.”
Ian Vanagas on Twitter / X
I've been enjoying writing screenshot essays lately because they provide more depth than tweets while being easier to finish and share than blogs.To help with this, I built a little app that makes it easier to make nice looking ones: https://t.co/mybEawgwgFEnjoy! https://t.co/gMHUCMzRbL— Ian Vanagas (@IanVanagas) March 30, 2024