Research on Semantic Analysis Algorithm of Network Public Opinion Based on LSTM

With the rapid development of information technology, social media platforms have become the main places for the public to discuss hot events, express opinions and emotions. Real time monitoring and effective analysis of online public opinion play a crucial role in grasping social dynamics, preventing and resolving social conflicts. This study aims to improve the accuracy of predicting changes in online public opinion by constructing and optimizing a semantic analysis algorithm based on LSTM, in order to more effectively manage and guide public emotions and maintain social stability. This article introduces the structures and functions of Recurrent Neural Networks (RNN), LSTM, and BiLSTM. Collect data through social media platforms and train and evaluate it using a BiLSTM model. The findings indicate that the BiLSTM model surpasses RNN and conventional LSTM models regarding predictive precision, particularly in managing long - range dependencies and grasping textual contextual details. The article also discusses the impact of Dropout technology and activation function selection on model performance. The BiLSTM model provides an effective method for predicting network public opinion trends, which helps to identify emotional outliers in a timely manner and prevent public opinion events from developing into major social events.