Research on Online Review Text Sentiment Analysis Using RoBERTa and BiLSTM Hybrid Method Based on Attention Mechanism

In the Internet era, the online review data of e-commerce platforms shows an explosive growth trend. The review data contains a large number of users’ personal views, hiding users’ emotional tendencies towards goods. Accurately capturing effective information in comments and mining users’ emotional tendencies is crucial for the development of e-commerce platforms. This article proposes a new online comment sentiment analysis model Ro2BLA. The model first embeds words into the input comment text through the RoBERTa layer, outputs dynamic word vectors with strong contextual relevance, and then inputs them into two layers of BiLSTM network layers for calculation, analyzing the deep bidirectional semantics in the comment text. Then, the Attention layer focuses on key information and extracts the deep hidden sentiment features contained in the text semantics. Finally, experiments are conducted on the Internet Movie Database (IMDB) and Amazon Fine Food Reviews data sets. By comparing different sentiment analysis models, the results show that the Ro2BLA model proposed in this article has better performance. The Ro2BLA model proposed in this article has reference significance for sentiment analysis of online comment texts on various platforms.