EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Explainable Emotion Recognition Using a Valence Arousal Dominance Calibrated IndoBERT for Indonesian Counseling Chatbots

Jimmy Agustian Loekito, Aris Tjahyanto, Rarasmaya Indraswari

Engineering, Technology & Applied Science Research · 06.06.2026 · 16 (3) · 35937-35949

DOI: 10.48084/etasr.17581 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering, Technology & Applied Science Research, online erschienen am 06.06.2026. Verfasst von Jimmy Agustian Loekito und 2 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Emotion recognition in text-based counseling is challenging as emotional cues are often subtle, implicit, and context-dependent. In contrast, conventional sentiment approaches reduce affect to coarse polarity and offer limited interpretability in clinical settings. This study proposes a Valence–Arousal–Dominance (VAD) Calibrated IndoBERT framework for explainable emotion recognition in Indonesian counseling dialogues. A total of 8,070 anonymized counseling dialogues, collected between 2021 and 2024 in accordance with institutional privacy regulations from an Indonesian university counseling center, were analyzed. The approach combines (i) fine-tuned IndoBERT for multi-class emotion classification, (ii) utterance-level VAD scoring using an amalgamated IndoVAD and NRC-VAD lexicon to provide affective coordinates, and (iii) a VAD rule layer with hysteresis that leverages dominance (perceived control) to disambiguate semantically similar high-arousal negative emotions (e.g., anger/fear/anxiety) and to reduce prediction instability across turns. The experimental results indicate that integrating VAD calibration and hysteresis improves Macro-F1 by 2.48%, reduces flip rate by 52%, and achieves an Expected Calibration Error (ECE) of 0.03 compared to an IndoBERT-only baseline. Three-dimensional VAD visualization further demonstrates improved class separability and interpretive stability, shifting emotion recognition from pure categorical labeling toward dimensional affect reasoning.

Zitieren

APA

Loekito, J. A., Tjahyanto, A., & Indraswari, R. (2026). Explainable Emotion Recognition Using a Valence Arousal Dominance Calibrated IndoBERT for Indonesian Counseling Chatbots. Engineering, Technology & Applied Science Research, 16(3), 35937-35949. https://doi.org/10.48084/etasr.17581

Vancouver

Loekito JA, Tjahyanto A, Indraswari R. Explainable Emotion Recognition Using a Valence Arousal Dominance Calibrated IndoBERT for Indonesian Counseling Chatbots. Engineering, Technology & Applied Science Research. 2026;16(3):35937-35949. doi:10.48084/etasr.17581

BibTeX
@article{Loekito2026,
  author = {Loekito, Jimmy Agustian and Tjahyanto, Aris and Indraswari, Rarasmaya},
  title = {Explainable Emotion Recognition Using a Valence Arousal Dominance Calibrated IndoBERT for Indonesian Counseling Chatbots},
  journal = {Engineering, Technology & Applied Science Research},
  year = {2026},
  volume = {16},
  number = {3},
  pages = {35937--35949},
  doi = {10.48084/etasr.17581}
}
Kostenlos starten →