Sentiment Analysis Method for Douban Movie Reviews Based on Prompt Learning

Movie reviews reflect the viewers evaluations of films. Conducting sentiment analysis on these reviews not only helps viewers select films they enjoy but also provides guidance for filmmakers in their creative processes. The sentiment analysis task for movie reviews has evolved from sentiment lexicon construction and machine learning to deep learning. These methods all rely on the fine-tuning paradigm, which has performance limitations when the downstream task objectives differ from the pretraining objectives. To address this issue, this paper adopts a prompt learning-based model. By designing task-description-based prompt templates, the downstream task is reformulated as a masked language prediction task, making full use of the semantic understanding of pre-trained language models. Experimental results show that compared to existing fine-tuning methods, the prompt learning-based approach improves accuracy by 6.8%-7.1% on the Douban movie review dataset, demonstrating the excellent performance of the prompt learning paradigm.