At the 2025 International AAAI Conference on Web and Social Media (ICWSM), held June 23-26 in Copenhagen, Denmark, researchers from USC Viterbi’s Information Sciences Institute (ISI) will present eleven papers exploring how online systems shape and are shaped by human behavior. Their work spans political polarization, platform bias, labor inequality, and crisis communication, using large-scale data from TikTok, Twitter/X, Reddit, and job platforms to better understand social vulnerability in the digital age.

Highlighted Research

Rethinking Which Jobs Are “Safe” from AI

Eun Cheol Choi, Qingyu Cao, Qi Guan, Shengzhu Peng, Po-Yuan Chen, Luca Luceri
Workshop on Data for the Wellbeing of Most Vulnerable

Think your job is safe from AI? The data tells a more complicated story. In Mapping Labor Market Vulnerability in the Age of AI, ISI researchers introduce a new way to measure job exposure to AI by linking real job postings with descriptions from AI-related patents. The study asks three key questions: Which sectors are most exposed? How do highly exposed jobs differ from others? And how does exposure relate to salaries? The results show wide variation across industries, with many of the most exposed roles found in lower-wage jobs that are disproportionately held by economically vulnerable workers. These findings suggest that AI may amplify existing labor market inequalities and raise important questions about how its impacts will be distributed.

Your Friends Might Be More Susceptible Than You Think

Luca Luceri, Jinyi Ye, Julie Jiang, Emilio Ferrara
Main Conference: Hidden Influence & Network Discovery

People often adopt behaviors they see others doing online, but not everyone is equally influenced. In The Susceptibility Paradox in Online Social Influence, ISI researchers explore how content spreads through social networks by comparing users who share something after seeing it from a friend with those who share it independently. The paper, which was selected by the Best Paper Selection committee as one of the best five papers of the conference, finds that users who are more easily influenced tend to be connected to others like them, creating clusters of highly susceptible individuals. This creates a surprising effect: most people are less influenceable than the average person in their network. The researchers call this the “susceptibility paradox” and suggest that this pattern may help explain how certain ideas or trends go viral, especially in tightly connected communities.

Why the Algorithm Rates German as More Aggressive

Gianluca Nogara, Francesco Pierri, Stefano Cresci, Luca Luceri, Petter Törnberg, Silvia Giordano
Main Conference: Bias and Fairness in Recommender Systems

Tools like Google’s Perspective API are widely used to detect toxic language, but they may not treat all languages equally. In Toxic Bias: Perspective API Misreads German as More Toxic, researchers show that the same messages are rated significantly more toxic in German than in English, even when the meaning is the same. The pattern holds across multiple datasets, topics, and translation methods. These findings raise serious concerns about how automated moderation tools perform in multilingual settings and whether they unfairly penalize some language communities. The paper calls for more transparent models that account for linguistic and cultural nuance.

What Social Media Reveals About the Abortion Divide

Ashwin Rao, Rong-Ching Chang, Qiankun Zhong, Kristina Lerman, Magdalena Wojcieszak
Main Conference: Online Polarization

After the U.S. Supreme Court overturned Roe v. Wade in June 2022, social media platforms saw a surge in abortion-related discussion. In Polarized Online Discourse on Abortion: Frames and Hostile Expressions among Liberals and Conservatives, ISI researchers examine more than 3.5 million tweets to understand how the debate unfolded across ideological lines. The study finds that liberals and conservatives use distinct rhetorical frames and tend to escalate hostility in response to each other’s messaging. Hostile expressions are especially common in quote tweets and replies, which are typically used to react directly to opposing views. The researchers also find that hostile language is often tied to moral framing, particularly among liberal users who emphasize rights and justice. Rather than simply reflecting disagreement, the study shows that hostility often emerges in reaction to perceived ideological opposition.

When Organized Crime Goes Viral: TikTok in Ecuador’s 2024 Political Crisis

Gabriela Pinto, Emilio Ferrara
Workshop on Data for the Wellbeing of Most Vulnerable

In early 2024, Ecuador faced a surge of political violence tied to organized crime, including gang-led prison riots and cartel-linked attacks. As events unfolded, TikTok became a central platform for sharing footage, commentary, and speculation. In A Multimodal TikTok Dataset of Ecuador’s 2024 Political Crisis and Organized Crime Discourse, researchers present a collection of more than 51,000 videos posted during the height of the crisis. Each video includes a transcript, sentiment label, and human-written summary, making the dataset usable for both computational and qualitative research. It offers a rare lens into how organized crime and political instability are framed and spread on a video-first platform, and is intended to support future work on misinformation, crisis narratives, and digital propaganda.

Full List of ISI-Affiliated Papers at ICWSM 2025

EDTok: A Dataset for Eating Disorder Content on TikTok
Charles Bickham, Bryan Ramirez-Gonzalez, Minh Duc Chu, Kristina Lerman, Emilio Ferrara

Fear and Loathing on the Frontline: Decoding the Language of Othering by Russia-Ukraine War Bloggers
Patrick Gerard, Tim Weninger, Kristina Lerman

Labeled Datasets for Research on Information Operations
Ozgur Can Seckin, Manita Pote, Alexander Nwala, Lake Yin, Luca Luceri, Alessandro Flammini, Filippo Menczer

Mapping Labor Market Vulnerability in the Age of AI: Evidence from Job Postings and Patent Data
Eun Cheol Choi, Qingyu Cao, Qi Guan, Shengzhu Peng, Po-Yuan Chen, Luca Luceri 

Modeling Information Narrative Evolution on Telegram During the Russia-Ukraine War
Patrick Gerard, Svitlana Volkova, Louis Penafiel, Kristina Lerman, Tim Weninger

A Multimodal TikTok Dataset of Ecuador’s 2024 Political Crisis and Organized Crime Discourse
Gabriela Pinto, Emilio Ferrara

The Peripatetic Hater: Predicting Movement Among Hate Subreddits
Daniel Hickey, Daniel M.T. Fessler, Matheus Schmitz, Kristina Lerman, Keith Burghardt

Polarized Online Discourse on Abortion: Frames and Hostile Expressions among Liberals and Conservatives
Ashwin Rao, Rong-Ching Chang, Qiankun Zhong, Kristina Lerman, Magdalena Wojcieszak

A Public Dataset Tracking Social Media Discourse about the 2024 US Presidential Election on Twitter/X
Ashwin Balasubramanian, Vito Zou, Hitesh Narayana, Christina You, Luca Luceri, Emilio Ferrara

The Susceptibility Paradox in Online Social Influence
Luca Luceri, Jinyi Ye, Julie Jiang, Emilio Ferrara

Toxic Bias: Perspective API Misreads German as More Toxic
Gianluca Nogara, Francesco Pierri, Stefano Cresci, Luca Luceri, Petter Törnberg, Silvia Giordano

Note: Every effort was made to include all ISI-affiliated papers at ICWSM25. If your paper was inadvertently left out, please let us know at [email protected] so the list can be updated.

Published on June 23rd, 2025

Last updated on June 23rd, 2025