Kevin Yang

Swarms of AI bots can sway people’s beliefs – threatening democracy A simulation shows that social media bots powered by today’s AI can infiltrate human networks on social media and influence what people believe.

This story via @theconversation.com weaves together research done with @yang3kc.bsky.social @baottruong.bsky.social and several others: 🧪 Swarms of AI bots can sway people's beliefs – threatening democracy

theconversation.com/swarms-of-ai...

Our latest paper in @science.org warns about malicious AI swarms, agents capable of adaptive influence campaigns at scale. We already observed some in the wild (picture). AI is a real threat to democracy.
#SciencePolicyForum #ScienceResearch 🧪
Paper: doi.org/10.1126/scie...

🚨 New working paper 🚨

Can LLMs with reasoning + web search reliably fact-check political claims?

We evaluated 15 models from OpenAI, Google, Meta, and DeepSeek on 6,000+ PolitiFact claims (2007–2024).

Short answer: Not reliably—unless you give them curated evidence.

arxiv.org/abs/2511.18749

🌍 Are you new to @icwsm.bsky.social or JQD:DM
and based in a low- or middle-income country?

💡 Submit a project idea, get matched with a mentor, present virtually at ICWSM'26, and prepare a submission for Sept 2026!
📢 Call icwsm.org/2026/submit....
🚀 Apply by Jan 15 forms.gle/A9GkJboP7qi3...

𝗗𝘂𝗻𝗰𝗮𝗻 𝗪𝗮𝘁𝘁𝘀 & 𝗦𝘁𝗲𝘃𝗲𝗻 𝗦𝘁𝗿𝗼𝗴𝗮𝘁𝘇 will give their first-ever 𝗷𝗼𝗶𝗻𝘁 𝗸𝗲𝘆𝗻𝗼𝘁𝗲 at NetSci 2026! Their groundbreaking work has shaped how we understand networks, & this session will be a highlight of NetSci’s 20th anniversary.
Call for abstracts: tinyurl.com/3tbj2v83
Call for satellites: tinyurl.com/42sru6kz

Congrats!! Though sad to see you leave the US.

DomainDemo: a dataset of domain-sharing activities among different demographic groups on Twitter - Scientific Data Scientific Data - DomainDemo: a dataset of domain-sharing activities among different demographic groups on Twitter

ICYMI, our DomainDemo dataset, which describes how different demographic groups share domains on Twitter, is now available to download!

📄 Data descriptor: doi.org/10.1038/s415...
📈 Interactive app to explore the data: domaindemo.info
💽 Dataset: doi.org/10.5281/zeno...

We are excited to share that this paper has been published in Scientific Data. The whole dataset is now available to download. Please see the paper for details!

Link: doi.org/10.1038/s415...

Now live: “Declining information quality under new platform governance” by Burak Özturan, Alexi Quintana-Mathé, Nir Grinberg, Katherine Ognyanova, and David Lazer. misinforeview.hks.harvard.edu/article/decl...

Claude Code is secretly an excellent data analysis agent How I vide code papers with it

Wrote a new blog post on my experience using Claude Code to do data analysis with lessons I learned. I think it's a paradigm shift. Link: open.substack.com/pub/yang3kc/...

Digital News Report 2025 The most comprehensive study of news consumption, covering 48 markets around the world.

Great report on the worldwide news landscape 2025. reutersinstitute.politics.ox.ac.uk/digital-news...

An astronomical theme is emerging for this great #ica25 session as I'm also looking forward to presenting one of the new tools from our own @internetobservatory.org.au. Please come and join us this afternoon as we chart the constellations of public communication 🔭🌠. Note this is in the GRAND Hyatt

Our introduction to the National Internet Observatory, a platform for collecting and sharing online activity data designed for researchers, is happening this afternoon at #ica25

Attending #ICA this week. Let’s catch up 😀

Congratulations!!!

Congrats!!!

A command I found particularly useful in data analysis: data_frame.sample(10). Repeat it a few times, and you can get a sense of what's in your data frame.

Search Google AI model directly: google.com/search?q={qu... . Key is the "udm=50" part, not sure what it means.

The age of mass data access is gone with the sunset of major social media APIs. DSA-mandated data access programs sparked optimism, but do they provide adequate data access? We conducted a survey (180) + interview (19) study & the preprint is available on arXiv: arxiv.org/abs/2505.09877

Thank you for the feedback!

Cool study!

Post-Post-API Age: Studying Digital Platforms in Scant Data Access Times Over the past decade, data provided by digital platforms has informed substantial research in HCI to understand online human interaction and communication. Following the closure of major social media ...

We document the researchers' desired data access and provide recommendations for platforms, researchers, and policymakers.

Podcast: notebooklm.google.com/notebook/d33...

With the amazing Kayo Mimizuka, @m-dot-brown.bsky.social, and @jolukito.bsky.social .

3/3

Specifically, we identify a number of obstacles researchers face 🔒:
- Lack of awareness/clarity
- Complex/burdensome application processes
- Arbitrary/opaque rejections
- Unreliable/low-quality infra and data
- Exclusion of non-academic and Global South researchers

2/3

[New WP] With the closure of major social media APIs and the new data access mandates under DSA, we enter what we call the "post-post-API" era. But have researchers obtained the data they need? Our recent survey (180) + interview (19) study suggests a stark reality.

🔗 arxiv.org/abs/2505.09877

1/3

We will be at #WebSci2025 next Tuesday to give a thorough introduction to NIO. Come and check it out if you are interested!

GitHub - yang3kc/llm_for_css: Guide on how to use LLMs for computational social science research Guide on how to use LLMs for computational social science research - yang3kc/llm_for_css

I'm releasing the v2.0 version of my guide on using LLM (mainly OpenAI models) for computational social science tasks. Major changes:
- Adopt the new Responses API.
- Better handling of the structured output.

Link: github.com/yang3kc/llm_...
Please star if you find it useful~

Congrats to @baottruong.bsky.social and @matthewdeverna.com !!!

Congrats!!!

Screenshot of table showing number of grants and total value terminated by NSF directorate. STEM Education directorate is highlighted, with 417 terminations totaling 322 million dollars. CSV of table: Directorate,No. Terminated,Total Value Biological Sciences,46,"$27,960,075" Computer and Information Science and Engineering,85,"$44,630,113" Engineering,94,"$33,472,485" Geosciences,52,"$36,454,740" Mathematical and Physical Sciences,30,"$6,418,312" Office of the Director,7,"$4,859,738" STEM Education,417,"$322,680,063" "Social, Behavioral and Economic Sciences",266,"$77,635,904" "Technology, Innovation and Partnerships",46,"$62,231,309"

UPDATE: We have data on >1000 terminated #NSF grants at grant-watch.us. Thanks to all who have been contributing data the past few days.

We've also added a "Reports" section with some analyses. For NSF, we see that the STEM education directorate has been absolutely pummeled.

Paper alert! 🚨 We investigate Bluesky’s journey from an invitation only platform with a few thousands of users to reaching 30 million users in terms of user activity and network growth.

📄 Full paper: arxiv.org/abs/2504.12902
💻 Codes: github.com/osome-iu/ris...
💾 Dataset: zenodo.org/records/1506...