Research

This research line develops computational methods for measuring emotion and social meaning in online communication. Across three connected papers, I use Affect Control Theory (ACT) to model how emotional states shift during text-based interactions, where body language and vocal cues are absent. The work extends affective dictionaries by mapping word and emoji embeddings into ACT’s evaluation–potency–activity space, enabling models to represent not only whether language is positive or negative, but also whether it signals power, agency, intensity, or social alignment.
The main artifact from this line is a visual pipeline showing how short messages, emojis, and social-event contexts are transformed into interpretable affective representations. Early work modeled emotional transitions in chatbot-style messaging and showed how emoji representation can enrich affective lexicons for online conversations. The later BERTNN framework advances this idea by using contextual transformer embeddings to estimate affective meanings for new concepts, supporting scalable, culturally adaptive analysis of sentiment and social dynamics.

The RFID collision-warning study reported collision prediction with less than 14% false alarms. In the Mindful Driving project, linked UVA coverage reported nearly 6% fuel-economy improvement, more than 23 gallons of fuel saved per vehicle, and 457 pounds of annual greenhouse-gas reductions per vehicle.

This work compares polling and financial contributions during the 2020 Democratic primaries and uses change-point analysis to identify moments when campaign trajectories shift, including shifts associated with debate performance and candidate support.

This research line studies how assumptions about expectation formation affect the stability and interpretation of macro-style dynamic systems. In the rational expectations literature, the determinacy condition is often used as a criterion for identifying unique stable solutions. My early work revisits this assumption and argues that determinacy alone can be a weak or insufficient criterion, especially when the mathematical structure of the model does not fully capture how agents form and revise expectations.

The related multi-agent work proposes an alternative way to model systems with linear rational expectations by representing decision makers as predictive agents rather than relying only on centralized equilibrium conditions. In this view, agents estimate future states, optimize their own actions, and interact through system-level feedback. A useful visual artifact for this line would show the contrast between a traditional rational expectations equation and the behind-the-scenes agent-based process that generates expectations, decisions, and stable system behavior.