Text sentiment recognition model based on BERT and dynamic ensemble selection

This study proposes a method for recognizing sentiment in Russian satirical literature using the BERT (Bidirectional Encoder Representations from Transformers) algorithm, based on deep learning technology. Satirical literature, characterized by complex language structures and implicit expressions, presents challenges for natural language processing (NLP) algorithms, as traditional sentiment analysis models struggle to capture its semantic nuances effectively. To address this, the study employs BERT to construct a sentiment recognition model specifically tailored for Russian satirical texts, leveraging pre-trained bidirectional deep neural networks to detect and interpret subtle emotional features within the text. During model training, extensive Russian satirical corpora were annotated and extracted to optimize BERT's semantic adaptability in Russian. Additionally, a specialized satire detection task was integrated to enhance the model's ability to classify implicit expressions of satire within the sentiment spectrum. Experimental results demonstrate that the BERT-based model outperforms traditional machine learning algorithms, such as SVM and LSTM models, across accuracy, recall, and F1-score metrics. This study not only expands the application of BERT in multilingual satirical sentiment recognition but also provides technical support for cross-linguistic satirical sentiment analysis, offering reference value for computer recognition of non-English satirical texts.