Intelligence Information System for Transformer-Based Sentiment Analysis
Вантажиться...
Дата
2026
Назва журналу
Номер ISSN
Назва тому
Видавець
Анотація
The paper presents an intelligent information system for transformer-based sentiment analysis focused on detecting gender bias in text classification models. The study emphasizes that transformer architectures may assign different sentiment scores to semantically identical sentences depending on gender-related words, which can negatively affect the fairness and objectivity of NLP systems.
The proposed system uses counterfactual text generation by creating male and female versions of the same sentence and comparing the sentiment scores produced by a DistilBERT-based model. The architecture includes modules for text preprocessing, counterfactual generation, sentiment classification, bias detection, and result interpretation.
Experimental evaluation demonstrated that the model exhibited gender-related differences in a significant number of analyzed cases, confirming the relevance of fairness auditing in sentiment analysis systems. The developed approach can be applied for ethical evaluation and monitoring of transformer-based NLP models.
Опис
Ключові слова
Бібліографічний опис
Mazurets O., Kuzmak K., Rovinsky A., Kadynska V. Intelligence Information System for Transformer-Based Sentiment Analysis. Proceedings of IV International Scientific and Practical Conference «Advanced Technologies in Scientific Research». May 13-15, 2026. Rotterdam, Netherlands. Pp. 398-403.