Optimizing LSTM Models for Tweet Sentiment Analysis: A Hyperparameter Study

This research investigates the effectiveness of using LSTM models for sentiment analysis of tweet data. By preprocessing the data and systematically optimizing the hyperparameters (e.g., number of LSTM units, dropout rate, batch size, and epoch), an LSTM model was constructed for tweet sentiment classification. In the test set, the model achieves 75% accuracy, verifying its feasibility in short text. Additionally, the experiments show that choosing the appropriate batch size and epoch number can improve the efficiency and stability of the model, which provides a reference for future research on sentiment analysis.