Research on Short Text Classification Based on BERT with Fusion of Key Features and Temporal Features

Short text classification is a challenging task in natural language processing (NLP). Because of its short length and sparse context, the traditional classification methods are limited. Pre-trained models such as BERT have improved the semantic representation ability, but they are still insufficient in capturing the key features and temporal dependencies of short texts. In this paper, we propose a hybrid model named BAL. By designing a novel feature fusion architecture, BERT, attention mechanism and LSTM are integrated, and global semantic representation and local temporal features are combined to improve short text classification performance. Experiments on AG News and SST-2 data sets demonstrate that BAL model achieves higher accuracy and robustness in classification tasks than baseline model. The contribution of this paper is to propose an efficient feature fusion framework and confirm its superiority in short text classification through experiments. This method overcomes the limitations of existing models in short text processing and has potential applications in social media analysis and public opinion monitoring.