EUVIMEDEUVIMED entdecken
Crossref

PREDICTION OF AN INDIVIDUAL’S EMOTIONAL STATE BASED ON TEXTUAL DATA USING BERT AND PAD MODELS

O. A. Kopylchak, I. Y. Kazymyra, M. T. Kravchuk

Ukrainian Journal of Information Technology · 2026 · 8 (1) · 8-15

DOI: 10.23939/ujit2026.01.008 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Ukrainian Journal of Information Technology, 2026. Verfasst von O. A. Kopylchak und 2 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper examines the problem of predicting a user’s multidimensional emotional state from textual records under conditions where most existing text-based approaches emphasize either categorical emotion recognition or coarse sentiment polarity, which limits the interpretability of broader affective assessment. The study aims to develop and experimentally validate a model that combines context-sensitive emotion detection based on a BERT classifier fine-tuned on the GoEmotions corpus with a hierarchical aggregation scheme (emotion-to-mood-to-polarity) and a quantitative representation of affect in the Pleasure-Arousal-Dominance space. The empirical basis includes short user-generated texts such as messages, comments, diary fragments, and reflective notes. To ensure compatibility with the English-language emotion corpus, the input text is normalized, translated into English when required, and segmented into paragraphs and sentences. Emotional labels are assigned at the level of contextual sentence windows. These outputs are then aggregated into paragraph-level emotion frequencies, weighted by emotion-specific PAD vectors, and transformed into dominant mood and overall text polarity. The results show that the proposed hierarchy provides a logically consistent transition from local emotional cues to a more stable affective profile. Evaluation on a held-out 10% subset of GoEmotions yielded macro-average Precision, Recall, and F1-score values of 0.518863, 0.485856, and 0.496837, respectively. The highest F1-scores were obtained for gratitude, amusement, love, and remorse, whereas grief, realization, disappointment, and nervousness remained the weakest classes. Scenario-based experiments further showed that the model distinguishes positive, negative, neutral, and mixed texts while preserving affective heterogeneity through PAD-based representation rather than forcing the input into a single categorical state. The contribution of the paper is to substantiate a PAD-oriented and hierarchically aggregated framework for interpretable emotional-state prediction from text, suitable for affective analytics, digital diary analysis, and wellbeing-oriented text assessment, and to demonstrate its potential as an analytical foundation for adaptive, user-centered intelligent systems.

Zitieren

APA

Kopylchak, O. A., Kazymyra, I. Y., & Kravchuk, M. T. (2026). PREDICTION OF AN INDIVIDUAL’S EMOTIONAL STATE BASED ON TEXTUAL DATA USING BERT AND PAD MODELS. Ukrainian Journal of Information Technology, 8(1), 8-15. https://doi.org/10.23939/ujit2026.01.008

Vancouver

Kopylchak OA, Kazymyra IY, Kravchuk MT. PREDICTION OF AN INDIVIDUAL’S EMOTIONAL STATE BASED ON TEXTUAL DATA USING BERT AND PAD MODELS. Ukrainian Journal of Information Technology. 2026;8(1):8-15. doi:10.23939/ujit2026.01.008

BibTeX
@article{Kopylchak2026,
  author = {Kopylchak, O. A. and Kazymyra, I. Y. and Kravchuk, M. T.},
  title = {PREDICTION OF AN INDIVIDUAL’S EMOTIONAL STATE BASED ON TEXTUAL DATA USING BERT AND PAD MODELS},
  journal = {Ukrainian Journal of Information Technology},
  year = {2026},
  volume = {8},
  number = {1},
  pages = {8--15},
  doi = {10.23939/ujit2026.01.008}
}
Kostenlos starten →