Affect2Act: Graph Attention Networks for Emotion-Informed Decision Making

Current affective computing systems focus on recognizing emotions but struggle to use them for reasoning, which limits human-AI interaction. We introduce Affect2Act, a graph attention model that represents emotions as interconnected nodes and learns context-dependent relationships. This structure enables flexible reasoning over emotional states. In a synthetic decision-making benchmark, Affect2Act achieves 87.67\% accuracy with an emotional balance of 65.47\%, slightly exceeding Graph Convolutional Networks (86.73\%) and clearly surpassing the attention-only (70.67\%) and MLP (45.67\%) baselines. These findings suggest that graph-structured emotional representations can support more robust, emotion-informed AI systems and open new directions for affective reasoning.