Sentiment Analysis Method for Network Public Opinion Text Based on ALBERT-CNN+BiLSTM-Attention Model

Sentiment analysis of network public opinion texts, which aims to extract public attitudes from massive textual data, has gained significant attention. While machine learning and deep learning methods are widely adopted, traditional approaches still face challenges such as manual feature dependency, the curse of dimensionality, and insufficient contextual semantic modeling. To address these limitations, this study proposes a hybrid model integrating ALBERT, CNN, BiLSTM, and attention mechanisms. Specifically, ALBERT generates text embeddings, CNN captures local features, BiLSTM extracts bidirectional contextual semantics, and the attention mechanism identifies sentiment-critical keywords. By synthesizing these features, the model achieves enhanced sentiment detection. Experiments on a Chinese Weibo sentiment dataset demonstrate the superiority of our method over four BERT-based baseline models, attaining an accuracy of 95.56%, precision of 95.18%, recall of 95.98%, and F1-score of 95.58%.