Авторы:
Издательство:
Дата издания
27.04.2026На каком языке издана
ISSN
19986688Область исследования
Аннотация (краткое описание)
The analysis of tonality in scientific texts, including citations, is actively advancing, enabling the identification of emotional coloring in references and their impact on scientific discourse. This study focuses on developing and evaluating a hybrid approach that integrates linguistic rules (analysis of parts of speech, syntactic dependencies, and negations) with machine learning algorithms (SVM, RF, NB, J48) to classify citation tonality. Experiments were conducted on the ACL Anthology (8700 sentences) and Clinical Trials (6500 additional sentences) corpora using stratified splitting (70/15/15 for train/val/test) and 5-fold cross-validation. The proposed method achieved 90% macro-F1 and 95% F1-score on the Athar dataset, and 85% macro-F1 on Clinical Trials, showing a 10–15% improvement over baseline models (BERT, LSTM). Ablation studies confirmed the contribution of linguistic rules (F1 increase of 5–7% when excluded). Statistical significance tests (McNemar, p<0.05) validated the robustness of the results. The approach proves effective for automated citation analysis and scientific impact assessment.DOI
10.55452/1998-6688-2026-23-1-52-67Тип публикаций
Количество цитирований
0Данный документ располагается в коллекциях
Международные публикации ученых SU
Scopus
Файлы документа
2499-6247-2-PB.pdf