UAlberta at SemEval-2026 Task 2: Temporal Fusion Models for Predicting Affect Over Time

AbstractWe describe our systems for the SemEval 2026 Task 2 on Predicting Variation in Emotional Valence and Arousal from Ecological Essays. To predict affect in a single instance, and for forecasting dispositional change, we use embeddings from a language model and a Recurrent Neural Network. To predict state changes from a previous timestep to the next, we use time-series forecasting. Our systems ranked first for forecasting dispositional change, and third for forecasting state change over time. We make our code publicly available.