Research on Text Classification Algorithm Based on Deep Learning

Short texts are characterized by their brevity, lack of standardization, and strong contextual dependency. Therefore, improving the accuracy of short text classification is a significant research topic. The use of deep learning models to effectively mine the semantic information of short texts for better text classification results warrants further exploration and research. In this paper, we propose a TextRCNN-ECA model based on the ECA attention mechanism to address the issue of weak feature extraction ability in the BiLSTM model. Firstly, our model utilizes BiLSTM to capture contextual feature information in the short text, which is then concatenated with the original word vectors to retain and control the transmission of contextual information effectively. Secondly, we employ max pooling to obtain the vector feature with the highest relevance to the text. Additionally, we introduce the ECA attention mechanism to assign weights to the text feature vectors, allowing for the identification of more critical features within the text. Experimental results demonstrate that TextRCNN-ECA exhibits a superior ability to capture key information in text.