A Weibo Sentiment Analysis Model Based on Attention Mechanisms and Deep Neural Networks

Sentiment Analysis has become a central research focus in the field of natural language processing in recent years, drawing significant attention from scholars. Weibo sentiment analysis can analyze the emotions expressed by users on Weibo through machine learning algorithms. This is helpful to understand the emotional state of college students and help ideological and political education. Traditional models for sentiment analysis face several issues, including neglecting edge information in texts, disruption of text sequence features by pooling layers, and shortcomings in feature extraction and key information identification. In order to further enhance the effectiveness of sentiment analysis, this paper proposes a Text Sentiment Analysis Model (AM DNN) for Sina Weibo using a dynamic deep neural network based on attention mechanisms and bidirectional gated recurrent unit (BiGRU) technology. To begin with, this paper employs wide convolutional kernels to extract edge features from the text and utilizes dynamic k-max pooling to preserve the sequential features of the text in relation to its position. Subsequently, a parallel hybrid structure of Deep Convolutional Neural Network (DCNN) and Bidirectional Gated Recurrent Unit (BiGRU) is established to mitigate the issue of partial feature loss, concurrently retaining local features and global context information, thereby enhancing the model’s feature extraction capabilities. Following the fusion of features, an attention mechanism is introduced in this paper to globalize its impact, thereby strengthening the model’s ability to identify key information. To validate the effectiveness of the proposed algorithm, the SMP2020-EWECT Weibo dataset is used for testing, encompassing six emotion categories such as joy, anger, and sadness. The AM DNN model proposed in this paper is compared with CNN, BiLSTM, Transformer, BERT models, BERT CNN, TextGCN, and BERT GCN. The results indicate that, considering the comprehensive evaluation metrics of accuracy, precision, recall, and F1 score, the improved AM DNN model outperforms other models, demonstrating superior classification performance in Weibo sentiment analysis. © 2024, Taiwan Ubiquitous Information CO LTD. All rights reserved.