APS
2026 APS Annual Convention · 2026
An Interpretable Deep Learning Model on Emotion Dynamics: Synthesizing Text and Numeric Data In Vector Autoregression Modeling
- Xinkai Du
University of Oslo - Sacha Epskamp
National University of Singapore - Johnny Zhang
University of Notre Dame
Abstract
We present a deep learning framework that unifies longitudinal text and Likert-scale data into a single vector autoregression (VAR) model to assess psychological dynamic systems. Combining variational autoencoders with a VAR(1) structure, it recovers interpretable latent factors and their temporal interdependencies, thereby bridging qualitative and quantitative designs in social science.
