UB researcher examines political communication processes in everchanging media ecosystem

How do humans make sense of rapidly evolving media technologies? Is AI unbiased? How does technology shape the way people talk online? How is data on social media gathered to inform our understanding of political climates?

For more than a decade, Yini Zhang, assistant professor in the Department of Communication, College of Arts and Sciences, has studied the evolving relationship between technology and democracy. As confusing as the world and political climate landscape is at the current moment, Zhang’s evidence-based approach to her research could not be more timely.

With expertise in social media, political communication and computational methods, her cutting-edge research sits at the forefront of a major shift within the communication landscape: the intense rise of artificial intelligence. Zhang is zeroing in on examining how AI is changing the information environment, and she’s putting those very AI tools to work to study the transformation.

She uses computational methods and content analysis while teaching computers to identify patterns in massive collections of social media posts. There are two approaches Zhang uses to compare her results: unsupervised and supervised machine leaning.

Unsupervised machine learning is just that: letting the computer produce results in a more unsupervised manner. In this approach, Zhang might feed millions of social media posts into an algorithm and ask it to discover patterns in them. For example, Zhang might feed 1 million tweets related to gun violence into a model and ask it to find patterns on its own, with no instructions and no categories set in advance. Human coders then step in to interpret what the algorithm found and check those results against manually coded data.

She also implements supervised machine learning, which requires a significant amount of human input to perform from the start. In this example, to identify posts expressing moral outrage in a political spectrum or pro or anti-gun control use, Zhang first develops a detailed codebook that gives every researcher on her team a shared definition for each construct. This allows them to work simultaneously with one another and hand code thousands of posts, reaching what researchers call “high inter-coder reliability.” This “ground truth” data trains the algorithm to recognize similar patterns across the larger dataset and allows it to work on bigger datasets from there. In other words, her team gets the information to a point where computers and humans more consistently agree on the parameters of the task and then continue to classify the content to be able to analyze large amounts in a manageable timeframe.

So, does AI perform better for her research? It depends.

“AI is highly accurate at identifying straightforward categories such social media posts that mention gun legislation in a specific state. It can decipher what information is correct or incorrect based on the law,” Zhang explains. “But more nuanced judgments, like whether a social media post frames gun violence as a systemic problem, are far more challenging for AI to measure and require human judgment,” she adds.

This human-centered process has always defined her methodology. “Human validation is the key. Large-scale results from machine learning and AI annotation always need the trained eye of a human expert,” she says. "AI is powerful, but humans will always be at the driver’s seat.”

She teaches a variety of undergraduate and graduate courses, including COM 205 Research Methods; COM 240 Introduction to Mass Communication; COM 337 Communication Theory; COM 485 Social Media and Society; COM 504 Research Methods; and COM 686 Social Media and Society. At the beginning of each semester, she introduces students to her work and offers them an opportunity to be a part of it. Since joining UB in 2020, three undergraduate students have worked in her lab alongside seven graduate students — three of whom have secured academic positions after graduation.

The Communication and Emerging Media Lab, which Zhang co-founded in fall 2021, is an independent collaborative space where students develop computational research skills while diversifying their ways of thinking. Students from UB and the University of Connecticut meet weekly over Zoom to conduct joint research and participate in workshops she leads on automated text analysis and other computational methods.

“It’s a space for students who are curious and they learn by doing,” Zhang says. “Students are often surprised by how difficult it is to achieve consistency in content coding. That difficulty is what makes it science.”

She stresses that media technologies are more complicated than we think. “What matters is the nuance and how we use them responsibly and steer them as individuals as well as a collective,” she says.

Zhang earned a PhD and MA from the University of Wisconsin–Madison, as well as an MA from Renmin University in China. She is frequently consulted as an expert by international media outlets, including The New York Times, BBC and USA Today.

In the fall 2026 semester, Zhang is being promoted to associate professor.