A personalized model for joint dialogue emotion classification and act recognition
Health Psychology Research · 25.05.2026 · 14 (2) · 0401
DOI: 10.36922/hpr.0401 ↗Kurz gefasst
Fachartikel in der Zeitschrift Health Psychology Research, online erschienen am 25.05.2026. Verfasst von Zhiqiang Ma und 4 weiteren Autor:innen.
Abstract (Originalsprache)
Background
Joint dialogue emotion classification and act recognition aim to simultaneously predict emotional and behavioral labels for each utterance. While existing models leverage discourse context, they largely overlook individual differences in sequential expression patterns.
Objective
This paper proposes a personalized model integrating personality for joint dialogue emotion classification and act recognition (PEDR-GAT) to incorporate personality traits into both local semantics and global temporal reasoning, thereby bridging inference gaps caused by neglected individual differences.
Methods
First, personality information and label distributions were used to initialize utterance representations, strengthening the association between personality-guided semantics and task labels. Second, graph attention mechanisms extracted global temporal dependencies across utterances and captured cross-task interactions. Finally, a co-attention mechanism integrated personality-guided local and global features to maintain long-term dependencies and supplement short-term contextual information.
Results
Experiments on Mastodon, DailyDialog, and CPED datasets show that PEDR-GAT outperforms baselines, achieving F1 improvements of 2.0%/0.6%, 2.3%/1.9%, and 5.3%/2.3% for emotion/act recognition, respectively. Personality-guided semantic label acquisition enhances semantics-label associations, and personality-fused graph attention effectively captures latent temporal patterns, demonstrating the value of incorporating individual differences into joint dialogue modeling.
Conclusion
PEDR-GAT effectively integrates personality traits into joint dialogue emotion and act recognition, demonstrating that modeling individual differences significantly enhances performance across multiple datasets.
Zitieren
APA
Ma, Z., Zhou, Y., Xu, B., Lv, K., & Liu, J. (2026). A personalized model for joint dialogue emotion classification and act recognition. Health Psychology Research, 14(2), 0401. https://doi.org/10.36922/hpr.0401
Vancouver
Ma Z, Zhou Y, Xu B, Lv K, Liu J. A personalized model for joint dialogue emotion classification and act recognition. Health Psychology Research. 2026;14(2):0401. doi:10.36922/hpr.0401
BibTeX
@article{Ma2026,
author = {Ma, Zhiqiang and Zhou, Yutong and Xu, Biqi and Lv, Kai and Liu, Jia},
title = {A personalized model for joint dialogue emotion classification and act recognition},
journal = {Health Psychology Research},
year = {2026},
volume = {14},
number = {2},
pages = {0401},
doi = {10.36922/hpr.0401}
}
EUVIMED