PVG at WASSA 2021: A Multi-Input, Multi-Task, Transformer-Based Architecture for Empathy and Distress Prediction

摘要: Active research pertaining to the affective phenomenon of empathy and distress is invaluable for improving human-machine interaction. Predicting intensities such complex emotions from textual data difficult, as these constructs are deeply rooted in psychological theory. Consequently, better prediction, it becomes imperative take into account ancillary factors test scores, demographic features, underlying latent primitive emotions, along with text’s undertone its complexity. This paper proffers team PVG’s solution WASSA 2021 Shared Task on Empathy Emotion Reaction News Stories. Leveraging data, score, intrinsic interdependencies empathy, we propose a multi-input, multi-task framework task score prediction. Here, prediction considered primary task, while emotion classification secondary auxiliary tasks. For system further boosted by addition lexical features. Our submission ranked 1st based average correlation (0.545) well (0.574), 2nd Pearson (0.517).