Emotion Analysis of Weibo Based on Long Short-Term Memory Neural Network

This paper proposes a sentiment analysis method based on the Long Short-Term Memory (LSTM) neural network, which achieves sentiment polarity classification of user comments on Weibo (a Chinese microblogging platform). The results are visualized, providing assistance in monitoring public opinion on Weibo. The study primarily employs both unidirectional and bidirectional LSTM models and compares their performance under different conditions through parallel repeated experiments. The model that performs better in the experiments is selected as the deep learning model. Additionally, Naive Bayes classifier, XGBoost classifier, and Support Vector Machine classifier are trained, and their optimal parameters are determined as machine learning models. The trained deep learning model and machine learning models are combined to form an ensemble model. The model is evaluated and tested using a test set and user comments on a specified topic, providing accuracy and top-5 accuracy metrics.