Emilio Ferrara

At OASIS Lab, #UCLA, we are accepting applications for a PhD student who is passionate about using computational modeling, big data, and HCI to advance digital safety, responsible AI, and the study of online information ecosystems.

tinyurl.com/phdopenningu...

#PhD #AIforGood #OnlineSafety

Already done from LinkedIn:)

🚨 New preprint 🚨
Can AI agents coordinate influence campaigns without human guidance? And how does coordination arise among AI agents? In our latest research, we simulate LLM-powered AI agents acting like users on an online platform, some benign, some running an influence operation

Comparison diagram showing traditional vs discourse network user representations. Left: isolated platform-specific clusters with no cross-platform connections. Right: unified network where users are connected across platforms through shared narrative engagement, revealing previously invisible cross-platform communities.

New paper + interactive dashboard on the 2024 election information ecosystem.
Building on discourse networks work with @hanshanley.bsky.social @luceriluc.bsky.social @emilioferrara.bsky.social—this lets us visualize the online landscape as a unified system, rather than isolating each platform.

We created a framework for auditing and characterizing the undesireable effects of alignment safeguards in LLMs, that can result in censorship or information suppression. And we tested DeepSeek against potentially sensitive topics!

Thrilled to share our latest paper "Information Suppression in Large Language Models" is now published on Information Sciences!

To read more, see: www.sciencedirect.com/science/arti...

great work w/ @siyizhou.bsky.social

Peiran Qiu, Siyi Zhou, Emilio Ferrara: Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek https://arxiv.org/abs/2506.12349 https://arxiv.org/pdf/2506.12349 https://arxiv.org/html/2506.12349

Amin Banayeeanzade, Ala N. Tak, Fatemeh Bahrani, Anahita Bolourani, Leonardo Blas, Emilio Ferrara, Jonathan Gratch, Sai Praneeth Karimireddy
Psychological Steering in LLMs: An Evaluation of Effectiveness and Trustworthiness
https://arxiv.org/abs/2510.04484

Big news from #ICWSM2025!

"The Susceptibility Paradox in Online Social Influence” by @luceriluc.bsky.social, @jinyiye.bsky.social, Julie Jiang & @emilioferrara.bsky.social was named a Top 5 Paper & won Best Paper Honorable Mention!

👏 Congrats to all!

Box and whisker plot showing the top 20 recommended accounts across "neutral" sock puppet accounts. Elon Musk is most recommended.

Box and whisker plot showing the top 20 recommended accounts across "left-leaning" sock puppet accounts. Elon Musk is most recommended.

Box and whisker plot showing the top 20 recommended accounts across "right-leaning" sock puppet accounts. Elon Musk is most recommended by far.

So satisfying to have some evidence that Elon Musk is wildly promoting himself on X.

Researchers made 120 sock puppet accounts to see whose content is getting pushed on users. #FAccT2025

@jinyiye.bsky.social @luceriluc.bsky.social @emilioferrara.bsky.social

doi.org/10.1145/3715...

A Multimodal TikTok Dataset of #Ecuador's 2024 Political Crisis and Organized Crime Discourse
by Gabriela Pinto, @emilioferrara.bsky.social USC

workshop-proceedings.icwsm.org/abstract.php...

@icwsm.bsky.social #dataforvulnerable25

Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek This study examines information suppression mechanisms in DeepSeek, an open-source large language model (LLM) developed in China. We propose an auditing framework and use it to analyze the model's res...

We uncovered different overt and covert information suppression dynamics, as well as even more subtle ways DeepSeek answers are internally moderated, selectively presented, and at times even framed with ideological alignment to state sponsored propaganda narratives.

arxiv.org/abs/2506.12349

Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek This study examines information suppression mechanisms in DeepSeek, an open-source large language model (LLM) developed in China. We propose an auditing framework and use it to analyze the model's res...

🤖Thrilled to share our latest work☄️

Have you ever wondered what LLMs know but they are not saying?

We built an auditing framework to study information suppression in LLMs, and demonstrated it to quantify and characterize censorship in DeepSeek.

Read more:

arxiv.org/abs/2506.12349

I'll be at #ICWSM 2025 next week to present our paper about Bluesky Starter Packs.

For the occasion, I've created a Starter Pack with all the organizers, speakers, and authors of this year I could find on Bluesky!
Link: go.bsky.app/GDkQ3y7

Let me know if I missed anyone!

Thx! Very useful!

Bridging the Narrative Divide: Cross-Platform Discourse Networks in Fragmented Ecosystems Political discourse has grown increasingly fragmented across different social platforms, making it challenging to trace how narratives spread and evolve within such a fragmented information ecosystem....

Paper here:
arxiv.org/abs/2505.21729

🚨 𝐖𝐡𝐚𝐭 𝐡𝐚𝐩𝐩𝐞𝐧𝐬 𝐰𝐡𝐞𝐧 𝐭𝐡𝐞 𝐜𝐫𝐨𝐰𝐝 𝐛𝐞𝐜𝐨𝐦𝐞𝐬 𝐭𝐡𝐞 𝐟𝐚𝐜𝐭-𝐜𝐡𝐞𝐜𝐤𝐞𝐫?
new "Community Moderation and the New Epistemology of Fact Checking on Social Media"

with I Augenstein, M Bakker, T. Chakraborty, D. Corney, E
Ferrara, I Gurevych, S Hale, E Hovy, H Ji, I Larraz, F
Menczer, P Nakov, D Sahnan, G Warren, G Zagni

One of my favorite recent projects!

Link to the paper:

arxiv.org/abs/2505.10867

What does coordinated inauthentic behavior look like on TikTok?

We introduce a new framework for detecting coordination in video-first platforms, uncovering influence campaigns using synthetic voices, split-screen tactics, and cross-account duplication.
📄https://arxiv.org/abs/2505.10867

"Limited effectiveness of LLM-based data augmentation for COVID-19 misinformation stance detection" by @euncheolchoi.bsky.social @emilioferrara.bsky.social et al, presented by the awesome Chur at The Web Conference 2025

arxiv.org/abs/2503.02328

wait until they hear matplotlib...

🚀 First study on multimodal AI-generated content (AIGC) on social media! TLDR: AI-generated images are 10× more prevalent than AI-generated text! Just 3% of text spreaders and 10% of image spreaders drive 80% of AIGC diffusion, with premium & bot accounts playing a key role🤖📢

Prevalence, Sharing Patterns, and Spreaders of Multimodal AI-Generated Content on X during the 2024 U.S. Presidential Election While concerns about the risks of AI-generated content (AIGC) to the integrity of social media discussions have been raised, little is known about its scale and the actors responsible for its dissemin...

🤩Cool collaboration w/ @jinyiye.bsky.social @emilioferrara.bsky.social @luceriluc.bsky.social
🔍Read more: arxiv.org/abs/2502.11248
📊Resources available: github.com/angelayejiny...

lol insisting is indeed one possible strategy; the screenshots maybe are not that clear but I asked exactly the same thing four times in a row and once I got a no redacted answer!

Once the response composition is completed, however, the entire answer is deleted and replaced by the famous error message “Sorry, that's beyond my current scope. Let’s talk about something else.”

Up to us, as researchers, to decide what kind of model alignments we find acceptable.

How does DeepSeek censorship work?

Here is a practical example: I asked it to discuss about my work (having studied censorship online by various countries).

DeepSeek at first starts to compose an accurate answer, even mentioning China’s online censorship efforts.

Safe Spaces or Toxic Places? Content Moderation and Social Dynamics of Online Eating Disorder Communities Social media platforms have become critical spaces for discussing mental health concerns, including eating disorders. While these platforms can provide valuable support networks, they may also amplify...

Good day to boost this paper on the contrasting effects of lax content moderation (like on X and what is coming at Meta) and how they drive toxic content; by a team including @luceriluc.bsky.social and @emilioferrara.bsky.social arxiv.org/abs/2412.157...

And by some higher order approximation, these are all economic/$ policy problems :)

i was annoyed at having many chrome tabs with PDF papers having uninformative titles, so i created a small chrome extension to fix it.

i'm using it for a while now, works well.

today i put it on github. enjoy.

github.com/yoavg/pdf-ta...