A Sentiment Analysis Model Based on Text Generation-OpnionSpanT5

ABSA is a type of fine-grained sentiment analysis, aiming to analyze the various sentiment types of a given sentence in more detail. Currently, more and more solutions adopt Seq2Seq-based architecture to perform sentiment analysis by outputting a sentence containing the required sentiment tuples. However, there is usually a problem, that is, when capturing opinion items, there may be a problem of missing extraction when facing opinion items with span. This paper proposes a model based on the paraphrase generation method, that is, before generating the quadruple, to extract opinion items separately. A BiLSTM-CRF layer is added in front of the PARA model to extract opinion items separately, and then the opinion items and the original sentences are input into the subsequent T5 model to generate emotional interpretations. We call it OpinionSpanT5, which is effective Fixed an error that occurred when generating opinion items.