Sentiment analysis in Chinese microblogs with long short-term memory and graph attention network

Sentiment analysis in microblogs is crucial for understanding public opinion and social trends, yet it faces challenges due to the brevity, informality, and complexity of microblog content. This study proposes an efficient sentiment analysis framework tailored for Chinese microblogs, combining Long Short-Term Memory (LSTM) networks and Graph Attention Networks (GAT) to address these challenges. The framework integrates data preprocessing, word embedding, and feature extraction through LSTM and GAT layers, followed by softmax-based classification. Experiments on a real-world Chinese microblog dataset demonstrate that the proposed method demonstrate state-of-the-art performance, with accuracy rates of 94.66% for binary classification and 85.68% for fine-grained classification, which outperform the baseline models of CNN, standalone LSTM, and LSTM+GCN by notable margins, clearly demonstrating the superiority of the LSTM-GAT integration over existing state-of-the-art approaches for Chinese microblog sentiment analysis.