Fine-grained Sentiment-Topic Analysis of Social Network Text Based on ERNIE and BiLSTM

The present work introduces a novel approach that combines pre-trained models with Bidirectional Long Short-Term Memory (BiLSTM) networks for the purpose of sentiment classification to six distinct categories. Fine-grained sentiment features are then employed in conjunction with Dynamic Topic Modelling (DTM) to monitor subjects and delve further into the peculiarities of public opinion. This study employed the fine-grained sentiment-topic analysis approach to compare the distribution of user sentiment and themes on international and Chinese platforms. The analysis focused on social media text data pertaining to nuclear wastewater releases in Japan. It gives novel perspectives on the management of public opinion concerning the global environment and provides useful references for improving international communication.