The 2025 ACM Web Conference, held April 28th to May 2nd in Sydney, Australia, has been the premier venue for web research since 1989, introducing groundbreaking technologies like Google’s search engine architecture and the YAGO knowledge base. 

This year, researchers from the USC Viterbi Information Sciences Institute (ISI) return to the conference to present a number of innovative papers. Their contributions span critical issues from detecting information operations on social media to enhancing smart city development. These papers demonstrate ISI’s continued leadership in shaping the future of the web through collaborative, interdisciplinary research that addresses today’s most pressing digital challenges.

Research Spotlights 2025

Exposing Cross-Platform Coordinated Inauthentic Activity in the Run-Up to the 2024 US Election
Federico Cinus, Marco Minici, Luca Luceri, Emilio Ferrara

Disinformation campaigns often operate across multiple social media channels, using a variety of techniques to manipulate public opinion and promote specific narratives. Detecting these adversarial efforts across the broader information ecosystem is crucial to safeguarding the integrity of online discourse, particularly during geopolitical events. While researchers have made progress in identifying such activity on individual platforms, there has been limited work focused on detecting coordinated campaigns that span multiple platforms.

Now, a new paper from ISI has introduced a methodological framework to identify these multi-platform campaigns. “This work seeks to uncover previously unseen operations both within and across diverse social media platforms and messaging apps,” said ISI Research Assistant Professor Luca Luceri, who supervised the study. 

The research team developed an advanced coordination detection model and applied it to newly collected data on online conversations related to the 2024 U.S. election. Their network-based model identified coordinated communities exhibiting suspicious sharing behaviors within and across platforms, and surfaced evidence of potential foreign interference in domestic political discourse.

In-Group Love, Out-Group Hate: A Framework to Measure Affective Polarization via Contentious Online Discussions
Buddhika Nettasinghe, Ashwin Rao, Bohan Jiang, Allon Percus, Kristina Lerman

In any healthy democracy, disagreements over policy are to be expected—and even welcomed. But when these disagreements become emotionally charged, it becomes harder to have constructive dialogue. This can undermine efforts to address public challenges effectively.

ISI computer scientists set out to understand what’s really driving this kind of emotional division—known as affective polarization—whether it’s more about rallying behind one’s own group, or disdain towards another. The research team, including ISI Research Assistant Ashwin Rao and Senior Principal Scientist Kristina Lerman alongside collaborators Buddhika Nettasinghe, Bohan Jiang, and Allon Percus, created a model that accurately captures decision-making within emotional polarized social networks and helps explain the rapid emergence of a partisan gap in attitudes towards contentious issues, including masking and lockdowns. 

“By quantifying these dynamics, we can begin to explore ways to mitigate polarization,” Rao, a research assistant at ISI, said. “Our aim is to contribute to solutions, and hopefully this conference helps spark the kinds of conversations and collaborations that make that possible.”

Embedding Spatial and Semantic Contexts for Geo-Entity Typing in Smart City Applications
Basel Shbita, Binh Vu, Fandel Lin, Craig A. Knoblock 

Geospatial data could transform urban development, governance, and public services in modern smart cities. However, the increasing availability of geospatial data presents a challenge: current AI geospatial applications lack robust, standardized, and automatic methods for classifying and understanding how smart cities interpret and utilize map data. 

Now, a new paper from ISI has introduced an AI system that can automatically classify geographic features by analyzing their shape, spatial characteristics, and neighborhood context. Their approach achieves remarkable 85% accuracy in identifying what different map elements represent—whether they’re buildings, roads, parks, or water features—using the community-built OpenStreetMap database. 

“By providing an interpretable classification of geo-entities, this research could help cities make smarter, faster decisions, whether it is in planning new parks, optimizing traffic, or responding to emergencies,” said Basel Shbita, a 2024 ISI Ph.D. graduate and paper first author. “The system offers the added benefit of improving over time thanks to its use of the ever-growing, openly accessible OpenStreetMap data ecosystem.”

Navigating Semantic Relations: Challenges for Language Models in Abstract Common-Sense Reasoning
Cole Gawin, Yidan Sun, Mayank Kejriwal

Do LLMs have the ability to “think” and truly understand the text they produce? Many prominent figures in the field of cognitive science argue that in order for a system to truly be “intelligent,” it must be grounded in commonsense knowledge that comes from the real world—this knowledge is argued to be the basis upon which abstract reasoning is made possible.

In a new paper from ISI, researchers explore these two fundamental constructs of human intelligence—commonsense knowledge and abstract reasoning—and whether or not LLMs are able to demonstrate them in similar ways to how humans can. “Our preliminary work in this paper shows that under many circumstances, the LLM performed poorly on tasks that a human would theoretically perform very well on,” said paper first author Cole Gawin, a USC student working under the supervision of ISI Principal Scientist Mayank Kejriwal. “This result adds to the growing body of evidence that LLMs are not truly intelligent systems, and cannot perform genuine abstract reasoning.” 

As LLMs become increasingly integrated into our lives and deployed in high-stakes domains, the paper’s findings could help developers, researchers, and policymakers remain critical of the current state of machine intelligence, Gawin said.

Unearthing a Billion Telegram Posts about the 2024 US Presidential Election: Development of a Public Dataset
Leonardo Blas, Luca Luceri, Emilio Ferrara

Telegram, a messaging app with lenient moderation policies, has been linked to a range of concerning activities—from the 2021 U.S. Capitol riot to illicit transactions and the spread of extremist content. 

To address this harmful behavior, ISI researchers have created the largest public Telegram dataset to date, accessing over 40,000 chats to collect over 1 billion messages related to the 2024 U.S. presidential election. The dataset offers an unprecedented opportunity for the research community and regulators to study political discussion on Telegram.

“Collecting Telegram data that captures these behaviors is the first step in developing social media tools to alleviate harmful online situations,” said Leonardo Blas, an ISI research assistant and paper first author. The team’s next steps include using the dataset for projects such as detecting protests and coordinated activity during the 2024 U.S. presidential election, and studying information diffusion, hate speech, and human trafficking in online platforms.

Complete list of accepted USC ISI papers

Limited Effectiveness of LLM-based Data Augmentation for COVID-19 Misinformation Stance Detection
Eun Cheol Choi, Ashwin Balasubramanian, Jinhu Qi, Emilio Ferrara

Moral Values Underpinning COVID-19 Online Communication Patterns
Julie Jiang, Luca Luceri, Emilio Ferrara

Network-informed Prompt Engineering against Organized Astroturf Campaigns under Extreme Class Imbalance
Nikos Kanakaris, Heng Ping, Xiongye Xiao, Nesreen K. Ahmed, Luca Luceri, Emilio Ferrara, Paul Bogdan

Beyond Interaction Patterns: Assessing Claims of Coordinated Inter-State Information Operations on Social Media
Valeria Pantè, David Axelrod, Alessandro Flammini, Filippo Menczer, Emilio Ferrara, Luca Luceri

Tracking the 2024 US Presidential Election Chatter on Tiktok: A Public Multimodal Dataset
Gabriela Pinto, Charles Bickham, Tanishq Salkar, Luca Luceri, Emilio Ferrara

Scalable Estimation of Exponential Random Graph Models on Large Networks
Yidan Sun

Published on April 28th, 2025

Last updated on May 7th, 2025