Attention-Enhanced BiLSTM with CNN Feature Fusion for Sentiment-Guided Social Media Noise Detection

Social media platforms contain large amounts of noisy and irrelevant content that reduce the quality of detecting sentiment and classifying text. This paper proposes Attention- Enhanced Sentiment-Guided Network for Noise Detection (ASGND), a combined deep learning model for social media noise detection. The model combines Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and an attention mechanism to capture local features, contextual information, and sentiment-related patterns from text data. Experiments were conducted on combined Twitter, Reddit, and Kaggle datasets containing 87,432 samples. Experimental results show that ASGND achieves an F1-score of 0.934 and exhibits better performance compared to manyconventional machine learning and existing deep learning models. Attention visualization shows that the focus of the model is on sentimentrelated words while distinguishing noisy and clean posts.Component removal analysis indicates that sentiment-guided attention contributes significantly to model performance. The proposed framework processes 1,200 posts per second on a single NVIDIA A100 GPU.The model can efficiently handle large volumes of social media data during analysis.