Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Sol Messing Profile picture
Sol Messing
@SolomonMg
Research Professor @NYU @CSMaP_NYU, @Georgetown MDI. Past: @Twitter, Acronym, @Facebook, @PewResearch, @Stanford.
Jun 18 • 22 tweets • 5 min read
🧵 NEW PAPER - "Quantifying Narrative Similarity Across Languages" is live at SMR. We combine NLP w pairwise LLM evals to track ideas across media ecosystems and languages. We also figured out a way to eval unsupervised probs on widely understood supervised performance metrics. Image This work was with my 1st co-author Hannah Waight (not on X) along with @AntonShirikov @mollyeroberts @Jonathan_Nagler @j_a_tucker @j___greenfield @m_dot_brown @aslett_kevin
Nov 17, 2024 • 13 tweets • 4 min read
🧵The study everyone here is talking about does NOT provide evidence that Twitter/X pushed a pro-Republican home timeline ranking change in July 2023.

The cascading, multiplicative effects of ranking changes likely explain the effect--details below Image 1. It's very likely that *something* in Twitter/X's home-timeline ranking system (HTRS) changed in July 2024. Image
Image
Apr 26, 2024 • 18 tweets • 4 min read
THREAD: TikTok's "algorithm" is magic and without it contemporary internet culture would simply melt away. Right?

What really happens if Douyin & TikTok split, leaving the latter without "the algorithm?" Image The media talks about TikTok's algorithm as if it's the company's secret sauce. Now, we don't technically know exactly how it works but here's the thing: these recommender systems, including Facebook, Instagram, and especially Reels, they are all quite similar under the hood.
Sep 30, 2023 • 25 tweets • 7 min read
1/ Many said Science went overboard on its cover for those Meta studies.

Science published my take yesterday.

It shows Meta’s algorithms actually tended NOT to increase ideological segregation in general, at least in 2020 🧵 Image 2/ Key point 1: González-Bailón et al 2023 claim newsfeed ranking increases ideological segregation (Fig 2B). BUT that’s based on domain-level analysis. URL-level analysis (Fig 2C) shows *no difference* in ideological segregation before and after feed-ranking. Image
Apr 3, 2023 • 29 tweets • 9 min read
I wrote about The Algorithm: using Musk's metrics in ship decisions, what the Republican/Democrat code means for democracy, how Twitter's API $ increase undermines transparency efforts, & on the tech bros claiming to analyze it 'so you can go viral.'

solomonmg.github.io/post/twitter-t… What does this mean for transparency? You need algorithmic audits to really understand what's happening on twitter, and with the recent API prince increases, to $500k/yr for meaningful access, made it incredibly difficult for research to audit this code in recent weeks
Jan 12, 2023 • 18 tweets • 7 min read
🚨MASSIVE NEW STUDY ON DIGITAL/FB POLITICAL AD EFFECTS 🚨 in @NatureHumBehav from Minali Aggarwal, @_JenAllen, @aecoppock, @dfrankow, Kelly Zhang, @jimmyeatcarbs, Andrew Beasly, Harry Hantman, Sylvan Zheng!

nature.com/articles/s4156…

Ungated: solomonmg.github.io/pdf/acronymNHB… Pundits and media commentators often assume large campaign effects, while many past studies find extremely small effects, often indistinguishable from zero. Measuring effects of the billions spent on political ads is one of the most significant challenges in the social sciences.
Oct 5, 2021 • 11 tweets • 2 min read
Important proposal from @persily. I was “the other side of the table” at Facebook when Nate was working on SS1, though we almost always agreed about the right way to share data. There’s a lot to like here for policymakers, researchers *and platforms* (brief thread) First, this straight up exempts university affiliated researchers from liability for scraping data for IRB-blessed projects. It's important to enshrine this into law to (1) protect researchers and (2) make it crystal clear that this work is normatively *good* for the world.
Sep 11, 2021 • 24 tweets • 8 min read
I have a few things to say about the @nytimes @daveyalba story on @SSOne FB Condor data error. It's a shame I wasn't interviewed First it sucks if you’ve written about the data. My heart goes out to you, and I have a sense of the work you have to do as a result. I understand there are engineers working to fix this ASAP.
Jun 10, 2021 • 19 tweets • 6 min read
Thread: @kmmunger et al's replication of work from @seanjwestwood & I brings up a key point for scholarship on tech: in this universe, we need to care not just about the person, but even more about the way the person interacts with the interface. This is exactly why Facebook’s algorithmically sorted News Feed has prompted such fierce public debate—what we read and how we read it are governed by the interface. Most obviously: we usually see what’s at the top of our feed and almost never see what’s at the bottom.
Feb 13, 2021 • 5 tweets • 4 min read
This week I started a new role at @Twitter, leading a team to work on what I'll call “social science for machine learning.” WERE HIRING.

We're tackling questions like... 1. How can survey data + ML be used together to create better experiences?

2. How can we build in safeguards to better understand algorithmic bias?

and ...
Dec 10, 2020 • 12 tweets • 4 min read
Thread: Democrats recently floated the idea of bringing back earmarks to break congressional gridlock. Will it work? Work from @justingrimmer @seanjwestwood & I does not suggest a rosy future. npr.org/2020/12/09/944… Earmarks started to really fall out of favor in 2008. In the debates, John McCain criticized Barack Obama for requesting "932 million in pork-barrel earmark spending" during while in the senate. Obama famously pledged to veto any bill that came to his desk with earmarks.
Nov 8, 2020 • 41 tweets • 13 min read
Post on polling in 2020. TLDR: projections based only on past voting returns better predicted 2020 results than poll averages in battleground states. I review how we used this to decide to spend in GA, then ask what might went wrong with polling in 2020? solomonmg.medium.com/past-vote-data… Back in April I had the notion to make some projections to see if it might make sense to expand where we at @anotheracronym might spend on mobilization in 2020.
Nov 7, 2020 • 4 tweets • 2 min read
Appreciate @StatModeling engaging w our work for a second time + this meme is one of the clearest findings from our paper—the margin of error didn’t affect certainty. At all. Actual result in the SM
Oct 1, 2020 • 14 tweets • 3 min read
Today we published an Op Ed in @USATODAY calling on journalists, policymakers & public intellectuals not to play to clickbait horse-race commentary in the 2020 election. That game led some to stay home in 2016 and there's too much at stake this time. usatoday.com/story/opinion/… Tuesday's debate raised the specter of voter intimidation, against a backdrop of pandemic-induced fear and social isolation. Many are trying to get by after losing a job, or helping anxious children get through distance learning.
Sep 28, 2020 • 4 tweets • 1 min read
If I worked for the Trump campaign I'd be stoked if a story like this hit & distracted everyone from Trump's insanely low tax rate + SCOTUS. Note that this leak almost certainly came *from someone on the right*. There is no mention of how this voter file was attained. There's only mention of a "leak."
Sep 22, 2020 • 8 tweets • 2 min read
DATA SCIENTISTS, GRAD STUDENTS: Wondering how you can put your skills to use for *this election* to help right the ship of American democracy? We need someone who can ramp up fast & help execute on: (1) analysis of field experiments, (2) analysis of survey experiments, & (3) other work in service of helping to improve a related ML product.
Sep 12, 2020 • 7 tweets • 2 min read
Let’s be clear this starts with far right talk radio, Sinclair, right wing cable news (OAN, NewsMax, FoxNews) and right wing websites. They spout propaganda + often misinformation to an old, conservative & nostalgic audience and make $ doing so. They give legitimacy to right wing conspiracy theories by providing the framing, narrative structure, and (incomplete + misleading) set of facts. Social media is a reflection of this.
Sep 3, 2020 • 18 tweets • 5 min read
THREAD: Facebook’s election announcement today may appear at first blush to help the left—which will benefit from higher turnout and a full count of mail-in ballots. But the story is more complicated & it may hurt the left in a few key ways. Facebook said: 1. No new ads in week before Nov 3

2. Claims people will get COVID if they vote will be nixed.

3. Content delegitimizing the election-vote by mail leads to fraud will be labeled

4. If Trump or Biden tries to declare victory before final count, likewise labeled
Aug 31, 2020 • 14 tweets • 4 min read
🚨 Huge new FB/Insta research initiative🚨 to finally conduct an independent, pre-registered research on the 2020 election. medium.com/@2020_election… This initiative, led by @TaliaStroud @j_a_tucker + @AnnieFranco @chadkdj on the FB side will study social media's impact on democracy with unprecedented data access.
Aug 29, 2020 • 10 tweets • 3 min read
This is effectively an analysis of engagement with public Facebook *pages*, not *content shared on FB*—a short thread from a research scientist who studies this & former Facebooker. Crowdtangle data describes *page posts and engagement therein*. But what matters much more are URLs shared with friends in personal posts, NOT reshares of page content.
Aug 9, 2020 • 4 tweets • 3 min read
Huge thanks to @MichaelBossetta for having me on @SMandPPodcast to talk about how the field of data science has changed since I got started in ~2006, building data sci teams @pewresearch & @anotheracronym, & the role of social science @Facebook socialmediaandpolitics.org I also talk about differential privacy & @SocSciOne.