赵紫东
I study how people interact with complex technology, and turn what I learn into metrics and product decisions.
I work across UX research and product data science. Most recently, I was a Senior Quantitative UX Researcher at Instagram, where I studied how people's experience of Reels connects to its ranking and recommendation backend, driving sentiment measurement strategy and helping improve content safety and personalization.
Previously, I was a research scientist at Yale's Science of Well-Being, where I built out its ecosystem of MOOC products and measured their efficacy. I hold a PhD in Psychology from Princeton.
My toolkit spans measurement, survey design, psychometrics, machine learning, and experimentation and causal inference. I also lean into qualitative methods like interviews and diary studies when they answer the question better.
Primary quantitative researcher on Reels Relevance, a three-person UXR team embedded with more than 10 data scientists and about 100 engineers. I guided the sentiment measurement strategy behind Reels ranking, prototyped an LLM classifier for contextually offensive content that secured funding and an engineering team for production testing, and built AI-native tools and workflows adopted by a research team of 12.
Led early-stage product and user research for a startup supporting children's social and emotional well-being, enabling several product pivots.
With Laurie Santos, shaped online well-being courses for parents and teens through efficacy research with more than 1.5 million learners, using survey, app, and lab data. Built pipelines and taught best practices so teams beyond research could run their own analyses, and managed three junior researchers.
PhD research in social psychology using surveys, experiments, neuroimaging, and computational modeling. Taught research methods, statistics, and programming for analytics.
Selected UX research case studies: the problem, the methods, and what changed.
In a previous life, I did experimental, computational, and neuroimaging research in academic psychology.
I did my PhD in the Princeton Social Neuroscience Lab, working mainly with Diana Tamir and Tom Griffiths. My dissertation, Dyadic Emotion Prediction: Knowledge Structure, Accuracy, and Implications, asked whether people can refine general social knowledge to make predictions about specific individuals, and what accuracy buys them.
Before that I earned a BA in Psychology at Vanderbilt, with minors in Quantitative Methods and Philosophy, doing clinical psychology research with Bunmi Olatunji and Sohee Park.
Our routine social lives are flush with information and demands. Navigating social situations requires more than passive reactions to the sights and sounds other people provide us. Rather, social savviness requires that we anticipate others' thoughts, feelings, and actions with reasonable accuracy. To predict well, we need to understand how people think and behave in general, as well as how a specific interaction partner might deviate from such patterns. The two lines of work below asked how people build these predictions, and how making them reshapes our own self-knowledge.
Despite wide variation across individuals, people typically learn to accurately tailor their predictions of how a specific familiar partner's thoughts and feelings might change over time. More remarkably, people manage to individuate their social predictions even for those with whom they only have very limited experiences. I combined probabilistic modeling and behavioral experiments to understand how structured mental representations might reflect statistical structures of the social world, seeking avenues by which we may rapidly extrapolate information about novel partners through inductive inference.
Memory is notoriously pliable. When memories are recalled, they become subject to alteration without us ever knowing. By extension, the self-knowledge we've built upon our memories is also constantly molded by a plethora of influences. This line of research focused on how the processes underlying social inferences can change our self-knowledge. When we interact with, or even just think about, other people, we often activate self-knowledge to anchor our inferences, and then adjust away from it by incorporating other information. In several collaborative projects, I investigated how our self-knowledge might be sculpted by such social anchoring-and-adjustment.
Meghan Meyer, Sasha Brietzke, Jordan Rubin-McGregor, and I showed that internally reasoning about others can change both our episodic and semantic self-knowledge to be more similar to our simulations of a social target. These changes can be extensive and lasting. Using fMRI and multivariate analyses, we investigated the neural underpinning of this phenomenon.
Madalina Vlasceanu, Elizabeth McDevitt, and I examined how memory might be shaped by conversations. When we converse, we base our anticipations of the partner's utterances on our own experiences. When our partner violates such expectations, it can actually weaken the pre-existing memories we drew upon. Using fMRI, we explored how the complex interplay between memory reinstatement, prediction error, and partner characteristics can shape our memory through social interactions.
Email is the best way to reach me:
zhaozidong1993 at gmail dot com
You can also find me on LinkedIn, GitHub, and Google Scholar.