The Research on an Online Review Sentiment Analysis Model Based on Improved RoBERTa

With the rapid advancement of internet technologies, platforms such as social media and e-commerce have deeply integrated into daily life, becoming indispensable channels for information exchange and opinion expression. Conducting in-depth sentiment analysis on the vast amount of online review data in a reasonable and effective manner not only allows for the accurate capture of public sentiment and trends in public opinion but also provides valuable insights for policymakers, aiding in informed decision-making. This approach helps to address practical issues in areas such as public opinion guidance and policy optimization. In response to the limitations of traditional sentiment classification models, which struggle with polysemy and inadequate feature extraction capabilities, this paper proposes an optimized sentiment analysis model based on the RoBERTa model. The proposed model integrates a pre-trained model with Bidirectional Gated Recurrent Units (BiGRU) and Attention (RoBERTa+BiGRU+Attention). The model achieves an accuracy of 95.36% on the IMDB dataset, and experimental results demonstrate that it can better capture and analyze the sentiment within review texts.