Utilizing Natural Language Processing for Efficient Text Analysis in the Era of Social Media

In this study, we explore the efficacy of various deep learning models, including TextCNN, Bi-GRU, and Attention-based LSTM, for short text classification tasks. We begin with data preprocessing and feature extraction techniques such as N -grams, information gain, and TF-IDF to prepare the datasets for analysis. The models are then trained and evaluated on both multi-category and four-category datasets. Our findings demonstrate that the LSTM model with an attention mechanism outperforms other models, achieving an accuracy of over $\mathbf{9 0 \%}$. We also provide a comprehensive performance evaluation using metrics like accuracy, precision, F1 score, confusion matrix, and ROC-AUC curve. The results suggest that integrating attention mechanisms significantly enhances the model’s ability to understand and classify short texts accurately.