Aravind Srinivasan Honored with Test of Time Award for Civil Unrest Forecasting

University of Maryland Professor of Computer Science and Distinguished University Professor Aravind Srinivasan has been recognized with a Test of Time award at the 2025 ACM Conference on Knowledge Discovery and Data Mining (KDD) for a paper on forecasting civil unrest using open source data.

Initially published in 2014, the paper—titled “Beating the News with EMBERS: Forecasting Civil Unrest using Open Source Indicators”—was coauthored by Srinivasan, his former Ph.D. student Khoa Trinh, now a researcher at Google, Naren Ramakrishnan and other collaborators. The project was part of a broader research initiative supported by the Intelligence Advanced Research Projects Activity (IARPA), focusing on predictive analytics in Latin America.

“This was a collaboration with faculty and researchers from multiple institutions,” Srinivasan said. “The goal was to use publicly available data, such as tweets, blogs, news stories and other indicators, to forecast events like civil unrest.”

The research team developed EMBERS, a system that utilizes machine learning models to analyze open-source data and generate probabilistic predictions. These predictions were evaluated in real time by an independent team at MITRE Corporation, with project support tied to the system’s forecasting performance.

While the project’s scope centered on Latin America, Srinivasan emphasized that the methods are generalizable to other regions and types of events. 

“This work falls under the broader umbrella of machine learning,” he said. “It touches on challenges such as data quality, causality and developing algorithms that can reason over temporal patterns.”

Srinivasan’s interest in machine learning dates back to the late 1990s, when he began exploring the relationship between data quantity and prediction accuracy alongside collaborator Dr. Philip Long. Over time, his research expanded from his core area of probabilistic algorithms to several real-world applications.

Reflecting on the award, Srinivasan said the recognition underscored the value of interdisciplinary collaboration and student contributions.

“I felt a lot of gratitude to my collaborators and especially to the students who carried out much of the ground-level work,” he said. “It was a strong interdisciplinary effort, and the training the students received was an important outcome.”

The EMBERS paper has been widely cited and remains influential in the field of applied data science. Its approach, leveraging publicly available information to generate real-time predictions, has relevance for a range of domains, from public health to financial systems.

However, Srinivasan also pointed to evolving challenges in the field, particularly with the increasing presence of synthetic or biased data online.

“Now, a lot of content on platforms like X is generated by bots or entities with specific agendas,” he said. “The more pressing question is how to determine which sources are trustworthy and what probability of correctness to assign to each.”

While algorithmic tools continue to improve, Srinivasan said advances in data verification will be essential to maintaining confidence in future machine learning systems.

“The techniques we developed still apply,” Srinivasan said. “The core challenges of working with real-world data—uncertainty, causality, temporal reasoning—continue to shape how we think about machine learning today.”

—Story by Samuel Malede Zewdu, CS Communications